Vous êtes sur la page 1sur 55

Deadly Embrace: Sovereign and Financial

Balance Sheets Doom Loops


Emmanuel Farhi

Jean Tirole

Harvard University

TSE

November 15, 2014

Abstract
The recent unravelling of the Eurozones financial integration raised concerns about
feedback loops between sovereign and banking insolvency, and provided an impetus
for the European banking union. This paper provides a double-decker bailout theory of the feedback loop that allows for both domestic bailouts of the banking system
by the domestic government and sovereign debt forgiveness by international creditors. Our theory has important implications for the re-nationalization of sovereign
debt, macroprudential regulation, and the rationale for banking unions.
Keywords: feedback loop, sovereign and corporate spreads, bailouts, sovereign default, strategic complementarities, debt maturity.
JEL numbers: F34, F36, F65, G28, H63.

Introduction

Rarely does an economic idea gather so wide a consensus as the evilness of the deadly
embrace, also called vicious circle or doom loop. The feedback loop between weak
bank balance sheets and sovereign fragility now faces almost universal opprobrium, from
The

research leading to these results has received funding from the European Research Council under
the European Communitys Seventh Framework Programme (FP7/2007-2013) Grant Agreement #249429.
Financial support of the research initiative "market risk and value creation" of the Chaire SCOR under
the aegis of the Fondation du Risque is also acknowledged. We thank seminar participants at numerous
institutions, as well as Xavier Freixas, Gita Gopinath, Olivier Jeanne, Anton Korinek, Guido Lorenzoni,
Thomas Philippon, and Vania Stravrakeva for useful comments and suggestions.

the IMF1 and central bankers to the entire political establishment and the European Commission, providing a major impetus to build the European banking union.
This paper seeks to analyze these developments by proposing a double-decker bailout
theory of the doom loop that allows for both domestic bailouts of the banking system by
the government and sovereign debt forgiveness by international creditors. The theory
has important implications for the re-nationalization of sovereign debt, macroprudential
regulation, and the rationale for a banking union (here defined as shared supervision).
Our model has three dates, 0, 1 and 2. At date 0, banks, which will need money
for their date-1 banking activities, manage their liquidity by holding domestic sovereign
bonds and foreign sovereign bonds (in the basic version of the model). Foreign sovereign
bonds are safe while domestic sovereign bonds are risky, and so the standard diversification argument would call for holding no domestic sovereign bonds.
Domestic sovereign bonds mature at date 2 in this basic version, and so the date-2
fiscal capability determines the sovereign spread. News accrues at date 1, that affects
the banks solvency (a financial shock) and/or the states date-2 fiscal capability (a fiscal
shock). A fiscal shock compounds the financial shock if banks hold government bonds.
While the government ex ante dislikes transferring resources to the banking sector, it cannot refrain from bailing out banks when facing the fait accompli of a banking liquidity
shortfall. The bailout further degrades the sovereigns ability to reimburse its debt at date
2, lowers the bond price and reduces bank solvency, etc., an amplification mechanism.
The multiplier reflecting the loss in sovereign bond price when a bailout is required increases with the extent of home bias.
We investigate the banks and the governments incentives to seek and prevent risk
taking, respectively. When banks can count on government bailouts, they optimally diversify as little as supervision allows them to, so as to enjoy the maximal put on taxpayer
money. Conversely, the government would like to limit risk and force diversification on
the banks. While holdings of domestic bonds are easily measured during a stress test,
they may not be so on a continuous basis ; furthermore, and mainly, banks may have
shrouded exposures to the domestic bond market though derivatives, guarantees or a
correlation of the banking book with domestic bonds. We accordingly assume that supervision is imperfect and study the extent to which banks are willing to incur costs so as to
evade diversification regulation.
In the process, we develop a new argument2 in favor of macro-prudential policies. The
1 See

e.g. Lagarde (2012).


arguments for going beyond the analysis of stand-alone bank solvency include the possibility of fire sales, interconnectedness and the policy response to, say, widespread maturity mismatches.
Neglected risk (Gennaioli et al. 2012) can also vindicate macroprudential policies.
2 Standard

consequences of individual banks undiversified portfolios, and therefore the desirability


of intense supervision, depend crucially on the other banks behaviors. We show that the
banks choices of opaqueness, and thereby their exposures to the sovereign, are strategic
complements: incurring the cost of making ones balance sheet more opaque is more
tempting if the put on taxpayer money is more attractive; in turn, this put is attractive
when the sovereign bond price is more volatile, which it is when the other banks take a
larger gamble. The corollaries to this insight are the existence of collective moral hazard
and the necessity of macroprudential regulation: The social cost of poor monitoring of
a banks domestic exposure is higher when other financial institutions are themselves
exposed. This is particularly true for institutions that the government is eager to rescue.
We then look at the incentives of foreign investors. We show that in bad states of nature, the legacy debt ends up on the wrong side of the Laffer curve once likely bailouts
and debt increases are factored in; investors thus have an incentive to forgive some debt.
This double-decker bailout in turn induces the government to turn a blind eye to undiversified bank portfolios. This however occurs only when the situation looks grim, a
prediction that fits well with the recent re-nationalization of government debt in the Eurozone. It also provides a new argument in favor of shared supervision. Indeed, if the
ex-post leniency of domestic supervisors is anticipated ex ante at the time of sovereign
debt issuance, then it is priced in the form of higher spreads. The government is better off
committing ex ante to a tough ex-post regulatory stance, but is tempted to relax it ex post.
If the government lacks commitment, then it benefits from relinquishing its supervisory
powers to a supranational supervisor.
Finally, we study five interesting extensions of the basic model. The first three do not
consider the possibility of debt forgiveness (either because debt is on the right side of the
legacy Laffer curve, or because debt is on the wrong side of the legacy Laffer curve but
investors have difficulties coordinating on a debt relief package).
First, we investigate the role of leverage by assuming that banks can seek refinancing
in markets at date 1. The feedback loop is then stronger, the higher the leverage. This is
especially so when sovereign defaults come together with defaults on banks private debt
contracts: As sovereign risk rises, banks have to reduce leverage because the probability
of a default on the private debt that they issue also rises. This requires a larger bailout,
which puts further pressure on the government budget etc. ad infinitum.
Second, there may be really adverse shocks for which the government can only undertake a partial bailout, as a full one would compromise public finances too much. We then
show that banks enter a rat race. While they wish to remain undiversified so as to enjoy the largest possible put on taxpayer money, they also try to jump ahead of the bailout
3

queue by being a bit more solvent, and therefore cheaper to rescue in the race for bailouts
in bad states of nature. Their holdings of foreign bonds are akin to bids in a first-price
auction, but the analysis is richer than the standard first-price auction in that the focus of
competition- the pot of subsidies to be distributed- depends on the distribution of bids,
namely the distribution of holdings of foreign bonds.
Third, we relax the assumption that sovereign debt maturity matches that of fiscal
capability. We compare our economy with long-term sovereign bonds which are claims
to coupons accruing at date 2 to an economy where sovereign bonds are short-term oneperiod bonds which are rolled over at date 1, assuming that the same amount is raised
at date 0. We show that a short maturity has both benefits and costs. The cost is that a
short maturity is bad for fiscal hedging. The benefit is that a short maturity reduces the
risk-shifting possibilities of banks. A short maturity is therefore an inefficient substitute
to regulation. As a result, a long maturity is preferable when supervision of bailout-prone
financial entities is efficient enough.
Fourth, we introduce foreign banks in the foreign (safe) country. Because of the bailout
guarantees, foreign banks also have an incentive to load up on risky domestic debt. The
foreign government has an incentive to supervise foreign banks so that they do not take
on too much domestic sovereign risk. The analysis then uncovers an additional rationale
for a banking union. Domestic supervision has positive external effects for the foreign
(safe) country. These effects are not internalized by the domestic government, and as
a result, supervision is too lax in the domestic economy. By transferring supervisory
decisions from the national to the international level, a banking union allows these effects
to be internalized, leading to a toughening of supervision in the domestic country and an
improvement of welfare.
Fifth, we relax the assumption that foreign assets are all safe. This relaxation is motivated by the multiplicity of troubled countries during the European crisis. Spanish banks,
say, could purchase Portuguese bonds and not only German ones. Therefore risk shifting
by Spanish banks could have occurred through the purchase of Portuguese bonds rather
than through a re-nationalization of the Spanish financial market. We therefore extend the
model to allow for multiple risky countries. We show that, provided that balance sheet
shocks and fiscal shocks within a country are at least slightly positively correlated and
that fiscal shocks across countries are imperfectly correlated (a reasonable assumption),
risk shifting solely through domestic bond holding is a strict equilibrium, implying that
the re-nationalization result is robust to multiple risky countries. We also show that, with
multiple risky countries, our double-decker bailout theory predicts that when the fiscal
outlook is bad, governments in risky countries have an incentive to relax supervision and
4

let their banks load up on risky domestic debt (and not risky foreign debt). All in all,
this extension shows that the multiple forces that we have identified for risk shifting in
the baseline model occur through the purchase of risky domestic debt rather than risky
foreign debt, implying that the re-nationalization results of the baseline model are robust
to multiple risky countries.
The paper is organized as follows: Section 2 sets up the framework and defines equilibrium. Section 3 identifies the sovereign and financial balance sheets feedback loop.
It derives the optimal regulation in the absence of renegotiation with sovereign debt investors; and it demonstrates that banks choices with respect to diversification are strategic complements. Section 4 explores the incentives of legacy creditors to engage in debt
forgiveness, how these incentives affect the regulatory stance of the government, and develops a rationale for a banking union. Section 5 presents the four extensions, concerning
the role of leverage and joint defaults, limited bailouts and endogenous diversification,
the maturity of sovereign debt, foreign banks in the foreign (safe) country, and multiple
risky countries. Finally, Section 6 summarizes the main insights and concludes.
Relationship to the literature.
Several papers have analyzed doom-loops and have
identified a feedback loop similar to the one described in our paper. In Acharya et al
(2013), the banks hold government bonds; the governments bailout of its financial sector
so as to preserve the latters lending to the non-financial sector reduces the financial sector credit risk in the short run, but increases the impact of future negative shocks to fiscal
capacity on the credit risk of the financial sector. As the papers title indicates, the stabilization of the financial sector is a Pyrrhic victory as it has deleterious long-term effects.
The theoretical model is a closed-economy model, in which default costs are internalized
by the government; it does not investigate topics such as re-nationalization, joint default,
and domestic vs. international regulation of banks that feature prominently in our analysis.
Cooper-Nikolov (2013) builds a model in which sovereign defaults are the outcome
of, as in Calvo (1988), self-fulfilling prophecies; similarly, banks fail because of DiamondDybvig (1983) runs. This model allows them to demonstrate the potential existence of
a doom loop: Worries about sovereign default generate concerns about the viability of
banks holding sovereign bonds; conversely, bank failures require bailouts, increasing the
volume of sovereign debt. The paper shows that equity cushions eliminate bad equilibria.
Our paper differs from Cooper-Nikolov in several respects; on the technical side, crises
are in our paper associated with fundamentals (although we of course find much interest
in Cooper-Nikolov self-fulfilling crises as well); this allows us to identify a multiplier and
5

to make unique predictions. Like Acharya et al, Cooper and Nikolov focus on a closed
economy.
Chari et al (2014) focus on a different aspect of bank holdings of government debt in a
closed economy. They show that forcing banks to hold a minimum fraction of their assets
in the form of government debt at or below market interest rates is dominated by other
fiscal instruments when the government can commit to a debt repayment policy, but can
lead to welfare improvements when the government lacks such commitment because it
increases credibility. There are no doom loops and instead a disciplining effect of bank
holdings of domestic debt on sovereign debt repayment. Moreover, the increased holdings of government debt by banks are driven by coercion of banks by the government.
Several recent contributions look at the contagion from sovereigns to banks in an open
economy, offering different hypotheses for sovereign debt re-nationalization and therefore sets of predictions and policy implications that differ from the unique ones summarized in Section 6; in this sense, our contribution is complementary with existing works.
The overall picture is the richness of the economics of interactions between sovereign and
bank solvency.
Broner et al (2013) consider environments in which the domestic government can default selectively on foreign investors. Selective default then makes domestic debt comparatively attractive to domestic residents in risky times, implying a re-nationalization.
In turn, increased domestic holdings of sovereign debt crowd out domestic banks investment in the real economy. The contagion channel (discrimination) differs from ours
(bailouts) and is one-way3 (from sovereign debt fragility to banks) rather than two-way;
so does the rationale for a banking union, as Broner et al view a union as a reduction
in discrimination between domestic and foreign investors while we focus on prudential
supervision.
Uhlig (2014)s model of financial repression, like ours, features banking supervision
and no discrimination among investors. It assumes a monetary union, whose central
bank is jointly backed by the member states and bails out commercial banks. Like in our
paper, a country may allow its banks to load up on domestic sovereign debt as the adverse
consequences will be shared abroad. Tolerating/ encouraging risk-shifting by banks that
have access to the unions central banks repurchase facility enables the risky country to
borrow more cheaply, a mechanism which bears some resemblance to our rationale for
strategic debt re-nationalization whereby governments allow domestic banks to buy up
their bonds in order to extract concessions from legacy creditors.
3A

two-way feedback loop arises in an extension of their model in which the cost of default is proportional to the amount of defaulted debt, with the proportion decreasing with the capital stock.

In Gennaioli et al (2014), domestic banks find domestic bonds attractive for a different reason than in our paper: the sovereigns internal cost of default (the drying-up of
domestic banks liquidity as there is neither discrimination nor bank bailouts) is high
when banking productivity is high; so sovereign repayment discipline is endogenously
positively correlated with the banks marginal utility of liquidity. This implies a renationalization of sovereign debt in bad times. There is no feedback loop but instead
a disciplining effect of bank holdings of domestic debt on sovereign debt repayment. Unlike our paper, the emphasis is not on prudential supervision and feedback loops.
Bocola (2014) emphasizes two different channels for the impact on banks of news regarding the possibility of a sovereign default. News that the government may default in
the future has adverse effects on the funding ability of exposed banks (liquidity channel),
and it raises the risk associated with lending to the productive sector (risk channel). There
is no feedback loop. In our model as in most aforementioned models that investigate the
links between banks and sovereigns, there is no risk channel.
An important ingredient of our analysis, as in Farhi-Tirole (2012) is that direct and
indirect exposures may be hard to assess, leading to supervisory failures, and that banks
will exploit the supervisory loopholes to secure cheaper financing and thereby increase
their return on equity. This ingredient is also shared by Mengus (2013a, b), who shows
that if furthermore banks in equilibrium (endogenously) choose heterogeneous portfolios, defaults involve an internal cost, and so a country may a) prefer not to default even
in the absence of sanctions, and b) may want to rescue another country despite the subsidy to third-party lenders to the defaulting country. The focus in Mengus is thus on
the impact of sovereign default on banks rather than on the doom-loop. Bolton-Jeanne
(2011) also study the international contagion of sovereign debt crises through the financial sector and their international fiscal implications. The focus in Bolton-Jeanne is on the
impact of sovereign default on banks and the role of banks in contagion rather than on
the doom-loop. Philippon-Skreta (2012) and Tirole (2012) study the design of government
interventions to rescue the financial sector in the presence of asymmetric information on
banks balance sheets. The focus is on ex-post interventions rather than on the doom loop.
In our model, the combination of limited commitment on the part of the government,
and ex-post bailouts gives rise to strategic complementarities in financial risk-taking, and
provides a rationale for macroprudential regulation. This occurs through a general equilibrium effect on the pricing of sovereign debt and the occurrence of default. This is an
important difference with other papers emphasizing strategic complementarities arising
from bailout guarantees, such as Schneider-Tornell (2004), Acharya-Yorulmazer (2008),
Ranciere et al. (2008), Diamond-Rajan (2012), Farhi-Tirole (2012), and Chari-Kehoe (2013),
7

which instead rely on mechanisms by which bailouts are extended only when sufficiently
many banks are in trouble.
In our model, ruling out bailouts, if possible, would be desirable. Bianchi (2013),
Stavrakeva (2013), and Keister (2014) argue that bailouts can have desirable properties
despite the associated moral hazard. In Bianchi (2013) and Stavrakeva (2013), this occurs
because bailouts help relax borrowing constraints in crises. In Keister (2014), this happens
because bailouts mitigate the incentives of depositors to run on banks in an environment
a la Diamond-Dybvig (1983). These papers stress that the optimal policy mix might involve bank bailouts combined with macroprudential policy. This possibility could arise
in our model, but we mostly focus on the case where it does not by assuming that banks
have enough net worth to take advantage of future investment opportunities provided
that they manage their liquidity prudently.

2
2.1

Model
Setup

We consider the following economy. There are three dates t {0, 1, 2} and a single good
at every date.
The economy is populated by international investors, a continuum of mass one of
domestic banking entrepreneurs and a continuum of mass one of domestic consumers. In
addition, there is a domestic government.
Uncertainty is gradually resolved over time. At date 1, a state of the world is realized
s S, with (full support) probability distribution d (s), where S is an interval of R+ . The
banking entrepreneurs balance sheets and the fiscal capacity of the government depend
on the realization of the state of the world s.
Private agents: international investors, banking entrepreneurs and consumers. International investors have a large endowment in every period. Their utility Vt = Et [2s=t cs ]
at date t is linear over consumption.
Consumers have a random endowment E [0, ) at date 2, with probability distribution function f ( E|s) and cumulative distribution function F ( E|s). The governments
only fiscal resources are at date 2: the government can tax the (random) endowment E of
domestic consumers. The endowment E can hence be interpreted as the fiscal capacity of
(1 F ( E|s)))
( f ( E|s)/(1 F ( E|s)))
0 and that ( f (E|s)/E
> 0.
the government. We assume that
s
The first inequality will imply that decreases in s are bad news for the fiscal capacity
8

of the government; the second is a monotone hazard rate condition that will imply a
quasi-concave Laffer curve. The two conditions are equivalent if s shifts the distribution
uniformly so that F ( E|s) = F ( E s). Consumers utility VtC = Et [c2C ] at date t is linear
over consumption at date 2. As usual, one can think of E as the consumers disposable
income beyond some incompressible level of consumption.
Banking entrepreneurs have an endowment A at date 0. At date 1, in state s, they have
a fixed-size investment opportunity which pays off at date 2. They can invest I (s) with a
payoff 1 (s) I (s) where 1 (s) > 1. The dependence of I and 1 on s more generally stands
dI (s)
for liquidity (or financial) shocks faced by banks. We assume that ds 0 so that low
s states are states in which banks badly need cash. The utility of banking entrepreneurs
VtB = Et [c2B ] at date t is linear over consumption at date 2.
We assume for the moment that the return from the investment project of banking entrepreneurs cannot be pledged to outside investors, and as result, banking entrepreneurs
cannot raise outside funding at date 1 (see Section 5.1 for a relaxation of this assumption).
Instead, they must self-finance the investment project I (s). Therefore, at date 0, banking
entrepreneurs trade their endowment A for financial assets (stores of value), part or all
of which they sell at date 1 to finance their investment project.4 We assume that A I
where I = maxsS I (s) so that if banking entrepreneurs manage or decide to preserve
their wealth between dates 0 and 1, they can always finance their investment project.
Assets. In the basic model, these financial assets are assumed to come in two forms,
domestic sovereign bonds in amount B0 , and foreign bonds. Both domestic and foreign
bonds are claims to a unit of good at date 2.
Except in Section 5.2, in which we will introduce competition among banks for access
to bailout funds, we look for a symmetric equilibrium, in which banks all choose the
same portfolio. We denote by b0 and b0 the representative banks holdings of domestic
sovereign bonds and foreign bonds. We assume that there are no short sales so that b0 0
and b0 0.
We assume that foreign bondswhich could be either private bonds or foreign government bondsare safe, and hence their price is always 1. By contrast, we assume that
domestic bonds are risky because the domestic government might default. We denote
their price in period 0 by p0 and their price in period 1 by p1 (s). We assume that p0 B0 > A
so that the marginal holder of domestic bonds is an international investor.
4 This

structure resembles Woodford (1990).

Welfare. At each point in time, the government evaluates welfare by subtracting default
costs (to be introduced below) from Wt = Et [c2C + B c2B + I (s)(s) I (s)], where Wt is a
weighted average of consumer welfare, banking entrepreneur welfare and investment
(s) I (s), where (s) is the mass of banking entrepreneurs who undertake their investment project. We assume that B < 1, and so pure consumption transfers to bankers are
costly.
The term I (s)(s) I (s) in the social welfare function captures an externality on other
agents in the economy, namely the welfare benefit for other banking stakeholders (borrowers, workers), from the banks ability to invest.5 This modeling of social preferences
thus allows for a wide range of preferences among economic agents.
Government debt, bailouts, defaults. The domestic government makes decisions sequentially, without commitment. At date 1, the government decides whether or not to
undertake a bailout of its domestic banks. At date 2, the government decides whether to
repay its debt or to default.
The government has some outstanding bonds B0 at date 0. We assume for the moment
that these bonds mature at date 2. In Section 5.3, we will investigate whether conclusions
are altered by a shorter maturity and whether the government optimally issues long-term
bonds.
At date 1, the government chooses whether or not to undertake a bailout of the financial sector. We assume that at date 1, the government inspects the balance sheets of
banks that apply for a bailout and so can, if it so desires, tailor individual bailout levels
to specific liquidity shortages of applying banks (which in equilibrium will end up being identical because of equilibrium symmetric portfolios).6 We denote by X (s) the total
transfer to the banks. In order to finance this transfer, the government must issue new
bonds B1 (s) B0 . The assumption that the government sets the amount it promises to reimburse, B1 (s), rather than the amount it borrows eliminates any multiplicity associated
5 Imagine

that, say, three categories of banking stakeholders benefit from the banks ability to invest.
First, and most obviously the banking entrepreneurs themselves: They receive 1 (s)(s) I (s), where 1 (s)
is the banks stake in continuation. Second, the higher (s) I (s), the better off their borrowers. Third, the
workers working in banks and industrial companies; to the extent that they are better off employed (e.g.,
they receive an efficiency wage) and that preserved employment is related to (s) I (s), then workers welfare grows with (s) I (s). Thus if 1F (s) and W
1 ( s ) denote the stakes of the industrial firms and the workers,
and if B , F and W denote the three categories of stakeholders welfare weights or political influence,
then B = B and I = ( F 1F (s) + W W
1 ( s )). This credit crunch interpretation can be formalized further along the lines of Holmstrm-Tirole (1997) (see Appendix A).
6 Alternatively, we could have followed Farhi and Tirole (2012) or Mengus (2013a, b) in assuming that
individual portfolios are imperfectly observed at the bailout date and that these portfolios are endogenously
heterogeneous. This would make bailouts more costly and the analysis more complex, without altering the
basic insights in our context.

10

Domestic debt
market clears at p0
(WTP of foreign
investors)
Supervisor chooses

b0 b0 ,
**

Banks select their


portfolios b0 , b0* b**
0
such that

State of nature s is realized,


determining fiscal prospects
f(E|s) and financial needs I(s).
Government issues B1(s)- B0 to
finance rescue package x(s).
Banks invest I(s) if they can.

2
Government
(non-selectively)
defaults iff
E < B1(s).

A= b0 + p0 b0* .

Figure 1: Timeline.
with erratic expectations as in Calvo (1988).
We assume that the weight I (s) on investment is high enough so that the government
always chooses to bail out the financial sector if such a bailout is needed, implying that
X (s) = max{ I (s) (b0 + p1 (s)b0 ), 0}. Finally, we assume that the government can always
raise enough funds at date 1 to finance the desired bailout (see Section 5.2 for a relaxation
of this assumption).
At date 2, the government decides whether or not to default on its debt. The government cannot discriminate between foreign and domestic bond holders, and hence cannot
selectively default on foreigners. The government incurs a fixed cost if it defaults on
its debt. We assume that the default cost is high enough, so that the government only
defaults if it cannot pay its debt, that is if and only if B1 (s) > E.7
Supervisory game. As we will note, banking bailouts provide banks with an incentive
to take risk. Conversely, and in the absence of sovereign debt forgiveness, the supervisor would like to limit risk taking. In general, an individual banks exposure to domestic bonds depends on a costly supervisory effort to detect hidden exposures and on the
banks costly effort to make these exposures opaque.
Rather than formalize the supervisory game in its entire generality we will assume that
there is an exogenous supervisory capability in the form of a minimum diversification
requirement, i.e. a lower bound on foreign holdings b0 . For a given supervisory capability b0 , the supervisor can then decide to be lenient by setting and effective minimum
sufficient condition is > E with probability one. A weaker sufficient condition is > B1 (s), but
involves the equilibrium object B1 (s).
7A

11

diversification requirement b0 b0 .
Except in Section 3.4, we assume that the supervisor can perfectly enforce the effective
minimum diversification requirement b0 , so that a banking entrepreneur must set b0
b0 . In Section 3.4, we assume instead that a banking entrepreneur can invest b0 in safe
foreign bonds and A b0 in domestic bonds at cost (b0 b0 ), where is a strictly
increasing and convex function on R with ( x ) = 0 for x 0.
Informational assumptions. We make the following informational assumptions. The
supervisory capability b0 and the amount of legacy debt B0 are publicly observable at
date 0. The decision regarding supervisory leniency b0 and the portfolios of banks are
not observable to international investors at date 0 but are publicly observable at date 1,
and so are the bailout X (s) and the amount of debt B1 (s).
Supervisory information is in general kept confidential. Why this is so is clear for
proprietary information embodied in the banks portfolio. But aggregate information is
no released either as regulators are also concerned by runs on an institution which is
assessed as risky by the supervisor. For example, in the US, CAMEL and RFI ratings
are not released to the public but only to the top management. The non-observability
assumption is therefore reasonable; but we have checked that the qualitative results also
hold in the slightly more complex case in which supervisory intensity is observed by the
market at date 0.
Illustrating example. We will make repeated use of the following simple example of
our more general setup. We assume that I (s) = I = A and 1 (s) = 1 for all s, and
that b0 A. The structure of uncertainty is as follows. With probability , the state s
is H and the endowment is high enough at E that there is no default. With probability
1 , the state is L and the endowment is high enough at E so that there is no default
with conditional probability x, intermediate e with conditional probability y, and 0 with
conditional probability 1 x y. In addition, we assume that e > B0 .
For E B1 ( L) > e, we have 1 F ( B1 ( L)| L) = x and so p1 ( L) = x and p0 = +
(1 ) x. For e B1 ( L) 0, we have 1 F ( B1 ( L)| L) = x + y and so p1 ( L) = x + y and
p0 = + (1 )( x + y). Depending on which of ( E B0 ) x and ( x + y)(e B0 ) is greater,
the level of debt B1 ( L) that maximizes revenue in state L is either E or e.

2.2

Equilibrium

In this section, we characterize the equilibrium of the model.

12

Bond prices and Laffer Curve. Because the marginal investor of domestic bonds is a
risk-neutral international investor, the prices of domestic bonds at dates 0 and 1 simply
reflect the relevant conditional default probability:
p1 (s) = 1 F ( B1 (s)|s),

p0 =
p1 (s)d (s).
At date 1 in state s, the government can thus collect ( B1 B0 )[1 F ( B1 |s)] by issuing B1 B0 . This revenue is strictly quasi-concave in B1 and increasing in s from our
assumptions on the distribution of the date-2 endowment E. It is always optimal for the
government to pick B1 = B1 (s) so as to be in the upward sloping part of the Laffer curve
in state s.
Banking entrepreneurs portfolios and supervision. Banking entrepreneurs invest their
net worth into foreign bonds b0 0 and domestic bonds b0 0 so that
A = b0 + p0 b0 .
At date 1, their pre-bailout net worth is b0 + p1 (s)b0 . If their pre-bailout net worth falls
short of the investment size I (s), they receive a government bailout. In a symmetric equilibrium, they receive X (s) = I (s) (b0 + p1 (s)b0 ). If their pre-bailout net worth exceeds
the investment size I (s), they simply save the difference by acquiring either domestic or
international bonds (at this stage, they are indifferent between both since they are risk
neutral over date-2 consumption).

Their expected utility is therefore V0B = [1 (s) I (s) + max{b0 + p1 (s)b0 I (s), 0}] d (s).

banking entrepreneurs always


Because p0 = p1 (s)d (s), I (s) is decreasing in s, A I,
choose b0 = b0 .8 This is intuitive: Banking entrepreneurs have an incentive to take as
much risk as possible to extract the biggest possible expected bailout from the government.
Bailouts and date-1 bond issuance. To finance the bailout at date 1 in state s
X (s) = max{ I (s) b0 ( A b0 )
8 Because

p1 ( s )
, 0}.
p0
p1 ( s )
p0 ) +
b00 > b0 . We

p1 (s) is increasing in s and I (s) is decreasing in s, there exists s such that b0 (1

p1 ( s )
p0 A

I (s) 0 if and only if s s. Note that if p1 (s) p0 , then s s. Now consider

p (s)
necessarily have s0 s. This implies that V0B0 V0B ss0 (b00 b0 )(1 1p0 )d (s) 0.

13

requires issuing B1 (s) B0 new bonds at date 1 with


p1 (s)[ B1 (s) B0 ] = X (s).
Date-1 debt B1 (s) B0 is the smallest solution of the following fixed-point equation

[ B1 (s) B0 ][1 F ( B1 (s)|s)] = max{ I (s) b0 ( A b0 )

1 F ( B1 (s)|s)
, 0}.
p0

(1)

The solution B1 (s) B0 is necessarily on the upward sloping part of the Laffer curve, and
we assume that equation (1) has a unique solution B1 (s) B0 on this upward sloping part
of the Laffer curve. If B1 (s) > B0 , this solution is then necessarily locally stable, by which
we mean that the slope of the left-hand side of (1) is greater than that of the right-hand
side.
There exists a cutoff s such that B1 (s) > B0 if s < s and B1 (s) = B0 for s s. FurtherdB1 (s)
1 (s)
more, we can show that ds
< 0 for s < s and similarly that dpds
> 0 for s < s, and for
( f ( E|s)/(1 F ( E|s)))
all s if
< 0 (strict inequality) .
s
Supervisory leniency. When setting the effective minimum diversification requirement
b0 b0 , the supervisor seeks to maximize welfare
"
W0 =

B1 (s)

[ E B1 (s)] f ( E|s)dE +

B1 (s)

[ E ] f ( E|s)dE + I (s) I (s) d (s)

1 (s) I (s) + max{b0

p (s)
+ ( A b0 ) 1
p0

I (s), 0} d (s),

taking p0 as given but taking into account the impact of b0 on p1 (s) and B1 (s) through
p1 (s) = 1 F ( B1 (s)|s),
p1 (s)[ B1 (s) B0 ] = max{ I (s) b0 ( A b0 )

p1 ( s )
, 0}.
p0

Sovereign and Financial Balance Sheets Feedback Loops

In this section, we illustrate the amplification mechanism arising from a feedback loop
between banks and sovereign balance sheets. We show that for a given supervisory capability, it is never optimal for the supervisor to engage in supervisory leniency. We then
14

characterize optimal first-best frictionless supervision (when the government can force
full diversification at no cost). We finally show that when supervision is imperfect, banks
domestic sovereign risk loadings are strategic complements, leading to the possibility of
multiple equilibria with varying degrees of banks domestic sovereign risk exposures,
and imparting a macroprudential dimension to supervision.

3.1

Amplification Mechanism

This feedback loop can be seen through the following fixed-point equation for the date-1
price of government bonds
p1 (s) = 1 F ( B1 (s)|s),
(2)
where
B1 (s) = B0 + max{

A b0
I (s) b0

, 0}.
p1 ( s )
p0

(3)

Using the implicit function theorem, we can then derive the following comparative static
result, assuming that a bailout occurs in state s, i.e. that s < s.
Proposition 1 (Feedback Loop). The sensitivity of date-1 bond prices p1 (s) to the state s when
a bailout is required is given by

dp1 (s)
=
ds

dI (s)
1
f ( B1 (s)|s) ds
p1 ( s )
.
I (s)b0
f
(
B
(
s
)|
s
)
1
p21 (s)

F ( B1 (s)|s)
s

(4)

The numerator encapsulates the direct effect of the change in s on the debt price p1 (s)
if there were no change in the price at which the government issues bonds to finance the
bailout and at which banking entrepreneurs liquidate their government bond holdings.
The first term in the numerator captures the direct change in the probability of no-default
at constant investment size I (s). The second term in the numerator captures the direct
impact of the change in the investment size I (s).
The denominator is positive because of the local stability of the selected fixed-point
solution to equations (2) and (3). It takes the form of a multiplier, which represents the
indirect effect of a change in s on the debt price p1 (s) through the change in the price at
which the government issues bonds and at which banking entrepreneurs liquidate their
government holdings. The multiplier is higher, the larger the amount of foreign-held
I (s)b
debt B1 (s) ( B0 b0 ) = p (s)0 that must be issued to finance the bailout (and hence
1
the higher the amount of domestic debt held by domestic banks, i.e. the lower is b0 ),
and the larger the semi-elasticity p 1(s) f ( B1 (s)|s) of the debt price p1 (s) to additional debt
1

15

issuances. This multiplier captures the feedback loop between banks and sovereigns as an
amplification mechanism: An increase in the default probability reduces the price p1 (s)
which increases the required bailout X (s) and hence the quantity of bonds B1 (s) B0 that
must be issued at date 1, which further reduces the price p1 (s) etc. ad infinitum.
dI (s)
Consider for example the case where ds = 0 so that there are no variation in in( f ( E|s)/(1 F ( E|s)))
vestment needs as we vary s, and assume that
< 0 (strict inequality).
s
Decreases in s are then just bad news for the fiscal capacity of the government. The effect
of a bad fiscal shock ds < 0 on bond prices p1 (s) is then amplified because some of these
bonds are held by the banking system, which increases the size of the required bailout,
worsening the fiscal problems etc. ad infinitum.
f ( E|s)
Similarly, consider the case where s = 0 so that there are no variations in fiscal
dI (s)
capacity as we vary s, and assume that ds < 0 (strict inequality). Decreases in s are then
just increases in the liquidity needs of entrepreneurs.9 Again, the effect of a bad financial
shock ds < 0 on bond prices p1 (s) is amplified because some of these bonds are held by
the banking system, which must then be bailed out, worsening the fiscal problems etc. ad
infinitum.

3.2

No Supervisory Leniency

In the model, it is never optimal for the supervisor to engage in supervisory leniency. In
other words, it is always optimal to set the effective minimum diversification requirement
b0 as high as allowed by supervisory capability b0 .
Proposition 2 (No Supervisory Leniency). It is never optimal for the supervisor to engage in
supervisory leniency so that b0 = b0 .
Proof. Using equations (2) and (3), we get

B1 (s)

[ E B1 (s)] f ( E|s)dE =

9 Although
d(1 (s) I (s))
ds

B1 (s)

[ E B0 ] f ( E|s)dE + min{b0 + ( A b0 )

p1 ( s )
I ( s ), 0}.
p0

this is not essential, in order for decreases in s to represent bad news, we also assume that

> 0, and

d( I (s) I (s))
ds

> 0.

16

We can use this equation to rewrite ex-ante welfare as


"
W0 =

B1 (s)

[ E B0 ] f ( E|s)dE +

B1 (s)

[ E ] f ( E|s)dE + I (s) I (s) d (s)

B [1 (s) I (s) + [ A I (s)]] d (s)

p (s)
+ (1 B ) min{b0 + ( A b0 ) 1
I (s), 0}d (s).
p0

The supervisor sets b0 b0 in order to maximize W0 taking p0 as given but subject to the
two constraints
p1 (s) = 1 F ( B1 (s)|s),
p1 (s)[ B1 (s) B0 ] = max{ I (s) b0 ( A b0 )

p1 ( s )
, 0}.
p0

The solution is clearly b0 = b0 .


In our environment, it is never optimal to engage in supervisory leniency because
of the inability of the government to commit not to bail out banks ex post. This creates a
standard soft budget constraint problem. The consequences of this problem are magnified
by the presence of the feedback loop between banks and their sovereigns.
In Section 4, we show that this conclusion can be overturned in the presence of debt
forgiveness. It can then be optimal to engage in supervisory leniency in order to extract
concessions from legacy creditors. We use the result established by Proposition 2 that
b0 = b0 throughout the paper except in Sections 4, as well as in the extensions in Sections
5.4 and 5.5 which also consider the possibility of debt forgiveness.

3.3

First-Best Frictionless Supervision

In this section, we investigate a related but different question. We perform a comparative


static with respect to supervisory capability b0 . As we argued earlier, the government may
have limited ability to force banks to diversify. Nonetheless, it is instructive to investigate
optimal supervision in the ideal theoretical situation where such limits to supervision
are absent. We refer to this situation as first-best, costless or frictionless supervision. The
occurrence of default is minimized when b0 = I so that banking entrepreneurs can always
finance their investment I (s) without requiring a bailout.
The welfare of banking entrepreneurs as well as total welfare are then independent of
the amount b0 b0 invested in foreign bonds above the floor b0 , where we have used
17

the fact that b0 = b0 . Reducing b0 below I on the other hand would reduce welfare for
two reasons. First, it would increase the occurrence of default. Second, it would have a
redistributive effect from consumers to banking entrepreneurs, which, as long as B < 1,
would be undesirable.
if feasible, maximizes
Proposition 3 (First-Best Frictionless Supervision). Setting b0 = I,
ex-ante welfare W0 .
there are no bailouts (B1 (s) = B0 for all s) and ex-ante welfare is given
Proof. For b0 = I,
by
"
W0 =

B0

[ E B0 ] f ( E|s)dE +

B0

#
I

[ E ] f ( E|s)dE + (s) I (s) d (s)

B [1 (s) I (s) + [ A I (s)]] d (s).

welfare can be rewritten as


For b0 < I,
"
W0 =

B1 (s)

[ E B0 ] f ( E|s)dE +

B1 (s)

[ E ] f ( E|s)dE + I (s) I (s) d (s)

B [1 (s) I (s) + [ A I (s)]] d (s)

p (s)
I (s), 0}d (s),
+ (1 B ) min{b0 + ( A b0 ) 1
p0

and is clearly strictly lower as long as the set of states s with B1 (s) > B0 has strictly
positive measure.
The optimal frictionless first-best supervision actually prevents the feedback loop from
occurring in the first place by prohibiting domestic banks from holding domestic sovereign
debt to an extent that could make them illiquid. We have already discussed in Section 2.1
creating the possisome reasons why we might observe suboptimal supervision b0 < I,
bility of the feedback loops that are the focus of this paper. These considerations lead us
to adopt a pragmatic position and treat b0 as a parameter.

3.4

Collective Moral Hazard

The rationale for liquidity regulation also has a macroprudential dimension. Indeed,
the benefits of liquidity regulation depend on the risk taken by the banking system as
18

a whole. For example, if for some reason only a fraction of banks take on domestic
sovereign debt, then the benefit from regulating the other banks is reduced (and might
even vanish) because the government has more fiscal space, reducing the riskiness of the
government bonds and hence the need for bailouts, and also weakening the feedback loop
between the remaining banks and the sovereign. We now show that for a given supervisory capacity, the incentives for banks to take on domestic sovereign debt are increased
when other banks do soa manifestation of the strategic complementarities in financial
risk-taking at work in the modelso that the effectiveness of supervision depends on the
risk taken by the banking system as a whole.
As discussed earlier, a banks ability to engage in risk taking depends not only on
supervisory policy, but also on its own ability to make its balance sheet opaque. Let us
capture this idea by taking the supervisory effort as a given and assume that each bank
indexed by i [0, 1] can (locally) select its individual level of foreign holdings b0 (i ) at
non-monetary cost (b0 b0 (i )), a strictly increasing and convex function with ( x ) = 0
for x 0. We look for a symmetric equilibrium in which all banks choose the same
b0 (i ) = b0 for all i. For simplicity, we focus on fiscal shocks and assume that I (s) = I
It is easy to see that the absence of
is independent of s. We also assume that A = I.

supervisory leniency b0 = b0 extends to the setting of this section. We therefore make


use of the fact that b0 = b0 throughout.
The banks choices of opaqueness, and thereby the exposures to the domestic government, are strategic complements: incurring the cost of making ones balance sheet more
opaque is more tempting if the put on taxpayer money is more attractive; in turn, this put
is attractive when the sovereign bond price is more volatile, which it is when the other
banks take a larger gamble.
To show this, note that for an individual bank i, given an aggregate b0 , the payoff from
investing b0 (i ) is

V0B (b0 (i ); b0 )

(s) +
1 (s) Id

( A b0 (i ))(

p1 ( s )
1)d (s) (b0 b0 (i )),
p0

where we have left the dependence of p0 and p1 (s) on b0 implicit.


Proposition 4 (Strategic Complementarities in Banks Domestic Exposures). Suppose that
There are strategic complementarities across banks
I (s) = I is independent of s, and that A = I.
V B (b (i );b )

in the choice of b0 (i ), i.e. the marginal benefit 0 b0 (i) 0 for a bank of increasing its individual
0
investment b0 (i ) in foreign bonds is increasing in the aggregate investment b0 of banks in foreign
bonds.

19

p1 ( s )
p0 .

For a given aggregate b0 , the random


1

variable e follows some distribution H (e) such that 0 (1 e)dH (e) = 1 (e 1)dH (e).
For an individual bank i, the payoff from investing b0 (i ) is
Proof. Denote by e the random variables

V0B (b0 (i ); b0 )

1 (s) Ad (s) +

( A b0 (i ))(e 1)dH (e) (b0 b0 (i )).

The marginal benefit of reducing b0 (i ) is given by


V B (b (i ); b0 )
0 0
=
b0 (i )

(e 1)dH (e) 0 (b0 b0 (i )).

Now consider two aggregate level b0 and b00 with b0 > b00 with associated prices p0 ,
p (s)
p1 (s), p00 , p10 (s) and distributions H and H 0 . Let s be such that 1p0 = 1 (and so bailouts
occur if and only if s < s). We proceed in two steps.
p0 (s)
p (s)
In the first step, we prove that p00 < p0 , p10 (s) = p1 (s) for s s, and 1p0 > 1p0
0
for s s. Indeed, the price p1 (s) is a locally stable solution of the following fixed-point
equation
1
1
p1 (s) = 1 F ( B0 + ( A b0 ) max{
, 0}|s).
p1 ( s )
p0
Towards a contradiction, suppose that p00 p0 . Then for any p1 (s), the right-hand side of
the above equation decreases when b0 is replaced by b00 . Hence p10 (s) < p1 (s) decreases
for all s, and strictly decreases for s < s. This contradicts the martingale property of
p (s)
p (s)
prices, and proves that p00 < p0 . For all s s, 1p0 > 1p0 1. Hence for all s s the
0

b00

p (s)

p (s)

pre-bailout net worth of banks satisfies


+ ( A b00 ) 1p0 > b0 + ( A b0 ) 1p0 . This in
0
turn implies that it is still the case that there are no bailouts for s > s when aggregate debt
is b00 . By implication, p10 (s) = p1 (s) is the same for s s.
In the second step, we use the first step to get

(e 1)dH (e)
1

p
( 0 e 1)dH (e) >
p 0

(e 1)dH (e).
1

The incentive to marginally reduce b0 (i ) is therefore higher when the aggregate foreign
debt level is b00 than when it is b0 :
V0B (b0 (i ); b0 )
V0B (b0 (i ); b00 )
>
.

b0 (i )
b0 (i )

20

As is well understood, depending on the exact shape of the cost function , these
strategic complementarities can lead to multiple equilibria: equilibria with low exposure
of domestic banks to domestic sovereign default risk (high b0 ) and equilibria with high
exposure of domestic banks to domestic sovereign default risk (low b0 ). This is the manifestation of a collective moral hazard problem as in Farhi-Tirole (2012).10 Because this is
not the focus of this paper, we simply illustrate this possibility with a simple example but
we do not develop this theme further.
Illustrating example. We consider the simple example introduced in Section 2.1. We
assume that ( E B0 ) x > ( x + y)(e B0 ) so that the revenue maximizing level of debt
(1 )(1 )
B1 ( L) in state L is E. We assume throughout that (0 )1 ( +(1 ) ) (0, b0 ) for
{ x, x + y}.
There are two possible equilibria depending on whether B1 ( L) e or B1 ( L) > e,
which determines the probability of repayment in state L. When B1 ( L) e, we have
= x + y, and when B1 ( L) > e, we have = x. And prices are given by p1 ( L) = ,
p0 = + (1 ).
The welfare of a banker i who invests b0 (i ) is
[1 A + ( A b0 (i ))(

(1 )(1 )
1
(b0 b0 (i )).
1)] + (1 )1 A = 1 A + ( A b0 (i ))
p0
+ (1 )

In order for banking entrepreneurs to choose b0 (0, b0 ), we must have


0 (b0 b0 (i )) =

(1 )(1 )
.
+ (1 )

The debt issuance condition is then


B1 ( L) = B0 + ( ),
where is a decreasing function defined by
( ) =

1 (1 )
(1 )(1 )
[ A b0 + (0 )1 (
)].
+ (1 )
+ (1 )

10 In

Farhi-Tirole (2012), we study a related model where the combination of limited commitment on
the part of the government, and ex-post untargeted bailouts gives rise to strategic complementarities in
financial risk-taking, and provides a rationale for macroprudential regulation. The main difference here is
that bailouts are perfectly targeted. Here there are also strategic complementarities in financial risk-taking,
which justify macroprudential regulation, but through a different, general equilibrium effect on the pricing
of debt and the occurrence of default.

21

We have an equilibrium with B1 ( L) e if and only if


( x + y) e B0 .

(5)

Similarly, we have an equilibrium with B1 ( L) > e if and only if


e B0 < ( x ) E B0 .

(6)

The two equilibria coexist if and only if


E B0 ( x ) > e B0 ( x + y).

(7)

Because the function is decreasing, we can always find values of B0 , E and e such that
condition (7) is verified so that there can be multiple equilibria for a range of parameter
values. These multiple equilibria are a consequence of the strategic complementarities in
the banks individual exposures to domestic sovereign default risk.
Proposition 5 (Multiple Equilibria). In the illustrating example, there are two possible equi (1 )(1 x )
libria. There is an equilibrium with low diversification b0 = b0 (0 )1 ( +(1 ) x ) and a
high probability of default (1 )(1 x ), which exists if and only if condition (5) is verified.
(1 )(1 x y)
There is also an equilibrium with high diversification b0 = b0 (0 )1 ( +(1 )( x+y) ) and a
low probability of default (1 )(1 x y), which exists if and only if condition (6) is verified.
The two equilibria coexist if and only if condition (7) is verified.
This example also has other interesting implications.
Proposition 6 (Multiple Equilibria and Debt Renationalization). In the illustrating example,
for B0 (0, E ( x )), the equilibrium with low diversification and high probability of default
is more likely to exist, the higher is legacy debt B0 and the lower is fiscal capacity (proxied by the
intermediate value of the endowment e). Conversely, the equilibrium with high diversification and
low probability of default is more likely to exist, the lower is legacy debt and the higher is fiscal
capacity.
This defines a precise sense in which high values of legacy debt or a reduction in fiscal capacity can lead to debt re-nationalization and offers a possible explanation for the
well-known fact that a re-nationalization of sovereign debt was observed in Europe as
the recent crisis intensified.11 Here this is due to the imperfect ability of the government
to limit the exposure of banks to domestic sovereign default risk through regulation. The
11 See

Broner et al (2013), Genaioli et al (2013) and Uhlig (2014) for careful documentations.

22

rationale for re-nationalization is based on the idea that sovereign bonds are more attractive to banks in bad times. But in bad times monitoring banks is also more attractive
to the supervisor. Proposition 6 nonetheless would still hold as long as the supervisory
capability does not adjust rapidly with the state of nature.
We return to this issue in Sections 4.2 and 5.2. In Section 4.2, we propose a different
mechanism for debt re-nationalization which relies on the desirability for the government
to allow banks to load up on domestic sovereign default risk in order to push legacy
creditors to forgive some debt, even if the government can perfectly regulate the banking
system. In Section 5.2, we uncover an opposing mechanism based on limits to the capacity
of the government to bail out the banking system.

Debt Forgiveness, Lax Supervision, and Banking Unions

In this section, we investigate the possibility of debt forgiveness at date 1. We show that
this can give rise to an incentive for lax supervision whereby the domestic government,
anticipating concessions from legacy creditors, turns a blind eye when its banks take on
domestic sovereign risk exposures and set b0 < b0 . If the ex-post leniency of domestic
supervisors is anticipated ex ante at the time of sovereign debt issuance, then it is priced
in the form of higher spreads. The government is better off committing ex ante to a tough
ex-post supervisory stance b0 = b0 , but is tempted to relax it ex post to b0 < b0 . If the
government lacks commitment, then it benefits from relinquishing its supervisory powers
to a supranational supervisor by joining a banking union.

4.1

Debt Forgiveness

We model date-1 debt forgiveness as follows. We assume that after the state of nature s
is observed at date 1, international investors can forgive some of the legacy debt to an
arbitrary B 0 B0 , before the government undertakes the bailout policy.
We show that it can be in the interest of legacy creditors (international investors who
Ab
hold the legacy debt B0 p0 0 ) to forgive some of the debt at date 1, bringing the overall
stock of legacy debt to B 0 B0 .12 In other words, there is a legacy Laffer curve, and
it is possible for legacy debt B0 to be on the wrong side of the legacy Laffer curve, i.e.
d( p1 (s; B 0 ) B 0 )
| B 0 = B0 < 0 where we have made the dependence of the date-1 price of debt
d B
0

12 Of

course organizing debt forgiveness requires coordination among legacy creditors to neutralize the
free-riding incentives of individual creditors.

23

p1 (s; B 0 ) on the post-debt forgiveness debt stock B 0 explicit.13 Moreover, we show that the
feedback loop between sovereign and financial balance sheets that we have characterized
in Section 3 makes it more likely that the economy is on the wrong side of the Laffer curve.
We can compute the sensitivity of the value p1 (s; B 0 ) B 0 of legacy debt with respect to
(post-debt forgiveness) legacy debt B 0 :
d[ p1 (s; B 0 ) B 0 ]
= p1 (s; B 0 ) B 0 [1
d B 0

Ab0
p0

B0

f ( B1 (s)|s)

]
1

I (s)b0
p21 (s; B 0 )

(8)

f ( B1 (s)|s)

The first term on the left-hand-side of equation (8) is the direct quantity-of-debt effect
of debt forgiveness. Because of this effect, marginal debt forgiveness d B 0 < 0 contributes negatively to the value p1 (s; B 0 ) B 0 of the claims of legacy creditors. The second
term on the left-hand-side of equation (8) is the indirect price-of-debt effect of forgiveness. Because of this effect, debt forgiveness d B 0 < 0 contributes positively to the value
p1 (s; B 0 ) B 0 of the claims of legacy creditors. The net effect of debt forgiveness depends
on the relative strength of these two effects.
The indirect price-of-debt effect of debt forgiveness is stronger, the more elastic is the
price p1 (s; B 0 ) to the amount of legacy debt B 0 . And the feedback loop between sovereign
and financial balance sheets that we have characterized in Section 3 works precisely to
increase this elasticity. Indeed, debt forgiveness increases the date-1 price of debt, which
improves the balance sheets of banking entrepreneurs, reducing the size of the bailout,
and hence reducing the need for the government to engage in additional borrowing at
date 1, which reduces the probability of default and further increases the date-1 price
of debt, etc. ad infinitum. The feedback loop therefore makes the price-of-debt effect
more potent, without affecting the quantity-of-debt effect, therefore pushing the economy
towards the decreasing part of the legacy Laffer curve p1 (s; B 0 ) B 0 .
When some debt forgiveness can improve the outcome of the legacy creditors, a mutually beneficial negotiation can take place between legacy creditors and the domestic
government. The outcome of the negotiation depends on the ability of legacy creditors to
coordinate and on the distribution of bargaining power between legacy creditors and the
domestic government. We assume that legacy creditors are able to coordinate, and have
all the bargaining power: They collectively make a take-it-or-leave-it offer to the domestic
government.
13 We

assume that banks do not free-ride on the renegotiation. All our results would go through if we
assumed instead that banks could perfectly free-ride on the renegotiation.

24

We can rewrite equation (8) as


1
d[ p1 (s; B 0 ) B 0 ]
= 1 B 0 [1
p1 (s; B 0 )
d B 0

Ab0
p0

B0

]
1

f ( B1 (s)|s)
1 F ( B1 (s)|s)
.
f ( B1 (s)|s)
I (s)b0
1 F ( B1 (s)|s) 1 F ( B1 (s)|s)

(9)

The best outcome that legacy creditors can achieve corresponds to the peak B0 (s) B0 of
the legacy Laffer curve
d[ p1 (s; B 0 ) B 0 ]
| B 0 = B0 (s) = 0.
(10)
d B 0
Proposition 7 (Legacy Laffer Curve and Debt Forgiveness). Suppose that there are only fiscal
shocks so that I (s) is independent of the state s. Then for every state s, the peak of the legacy Laffer
curve B0 (s) B0 is increasing in s so that worse states are associated with more debt forgiveness.
Proof. We can rewrite equation (10) as
I (s)b
f ( p (s;B (0s))
0
1

1 F(

+ B0 (s)(1

I (s)b0
p1 (s;B0 (s))

Ab0
p0

+ B0 (s)(1

B0

)|s)

Ab0
p0

B0

)|s)

I (s) b0

1
1 F(

I (s)b0
p1 (s;B0 (s))

B0 (s)(1

+ B0 (s)(1

Ab0
p0

B0

Ab0
p0

B0

)=

I (s)b
f ( p (s;B (0s))
0
1

)|s) 1 F (

+ B0 (s)(1

I (s)b0
p1 (s;B0 (s))

Ab0
p0

+ B0 (s)(1

B0

)|s)

Ab0
p0

B0

. (11)

)|s)

When I (s) is independent of s, the left-hand side of this equation is increasing in B0 (s)
while the right-hand side is decreasing in B0 (s). Hence the equation has a unique solution in B0 (s) which characterizes corresponds to the global maximum of the legacy Laffer
curve p1 (s; B0 (s)) B0 (s). The result follows easily from the monotone hazard rate assumption.

4.2

Strategic Supervisory Leniency

The possibility of a legacy Laffer curve can make it optimal for the government to engage
in supervisory leniency b0 < b0 , because it allows to extract larger concessions from
legacy creditors. Another way to put this is that the government might have incentives to
let its domestic banks load up on domestic sovereign debt in order to extract concessions
from legacy creditors. We illustrate this possibility with a simple example.

25

Illustrating example. We consider the simple example introduced in Section 2.1. Recall
We assume that e(1 +
that in this example, I (s) = I is independent of s and A = I.
y
I
I
x ) > B0 > e and that ( s ) = is independent of s. We assume that supervisory
+(1 ) x
capability is not too low A (1 x) [e( x + y) xB0 ] b0 . We now proceed to construct
an equilibrium where it is optimal for the government b0 < b0 so as to obtain concessions
from legacy creditors.
There is no debt forgiveness in state H and no default. At date 1, in state L, legacy
creditors either forgive no debt so that B0 ( L) = B0 or forgive debt B0 ( L) < B0 in the
following amount
[1 B0B(0L) xp+0y ]( A b0 )
= e,
(12)
B0 ( L) +
x+y
in which case B1 ( L) = e. There is debt forgiveness provided that when B0 ( L) is defined
by equation (12), the following condition is verified:

( x + y) B0 ( L) xB0 .

(13)

To summarize, using equations (12) and (13), there is debt forgiveness in state L if

(1 px0 )( A b0 )
x
B0 +
e.
x+y
x+y

(14)

In order to maximize welfare


W0 = ( E B0 ) + (1 )[ x ( E e) (1 x y)] + I A

+ B [1 A + ( A b0 )(

1
1)]. (15)
p0

It is then always optimal for the government to choose at date 0 the lowest value of b0
that satisfies equation (14). This defines an increasing function
b0 ( p0 ) = A

x+y
x
B0 ).
x (e
1 p0
x+y

The date-0 price is then given by p0 = + (1 ) x.14 And the equilibrium effective
diversification requirement is then b0 = b0 ( + (1 ) x ) which is guaranteed to be
less than b0 by our assumption that supervisory capability is not too low.
14 We

have p0 = + (1 )( x + y)

B0 ( L)
B0 ,

which using ( x + y) B0 ( L) = xB0 implies p0 = + (1 ) x.

26

Proposition 8 (Strategic Supervisory Leniency). In the illustrating example, it is optimal


for the government to engage in strategic supervisory leniency by setting b0 = b0 ( + (1
) x ) < b0 . The equilibrium effective diversification requirement b0 is decreasing in the probability 1 of the occurrence of the bad fiscal shock (state L).
Proposition 8 shows that in the illustrating example, it is optimal for the government
to engage in supervisory leniency b0 < b0 and allow domestic banks to take on more
domestic debt, and risk needing a bailout when the government experiences a bad fiscal
shock. This allows the government to extract more concessions from legacy creditors.
The government reduces the effective diversification requirement (lowers b0 ) when the
probability 1 of a bad fiscal shock where a debt renegotiation takes place increases,
because it makes it more attractive to extract concessions from legacy creditors.
This offers a possible explanation for the well-known fact that a re-nationalization of
sovereign debt was observed in Europe as the recent crisis intensified. In Section 3.4, we
proposed a different mechanism based on the imperfect ability of the government to limit
the exposure of banks to domestic sovereign default risk through supervision.
Remark. Excessively lax supervision can of course pre-date the crisis. Indeed, our formalism allows for agency costs within government, as the supervisor may put excessive
(relative to the population) weight on bankers welfare or too much weight on real estate
lending for instance. Proposition 8 then means that the prospect of debt forgiveness may
make supervisors even more lenient than they would be otherwise.

4.3

A Rationale for a Banking Union

In the illustrating example developed in Section 4.2, foreign investors are made worse off
by the relaxation of supervision of domestic banks by the domestic government: Once
they have lent, their welfare is maximized by a tough supervision b0 = b0 . Of course
their welfare is adversely impacted only if this relaxation of supervision is not anticipated
at the time of the debt issuance, otherwise it is fully priced in. Interestingly, in this latter
case, domestic welfare can be increased by a tough supervision b0 = b0 because of its
positive effects on the issuance price of date-0 debt. But this requires commitment on
the part of the domestic government not to relax supervision after the debt is issued.
A banking (in the sense of shared supervision) union can help deliver the commitment
outcome.
Indeed, building on the illustrating example of Section 4.2, consider the debt level
B 0 < B0 that generates the same amount of revenue at date 0 when the effective diversification requirement is b0 = b0 as the debt level B0 when the effective diversification
27

requirement is b0 = b0 ( + (1 ) x ). This debt level is defined implicitly by the equation

[ + (1 ) x ] B0 = p0 (b0 ; , B 0 ) B 0 ,
where we assume that the solution of this equation satisfies e(1 + rt ) > B 0 > e, and where
p0 (b0 ; , B 0 ) denotes the date-0 price when the effective diversification requirement is
b0 = b0 .15,16,17,18
0 is given by
The associated level of date-0 welfare W
0 = ( E B 0 ) + (1 )[ x ( E e) (1 x y)] + I A
W

+ B [1 A + ( A b0 )(

1
p0 (b0 ; , B 0 )

1)].

which is guaranteed to be greater than W0 .19


This requires commitment on the part of the domestic government since once the date0 debt B 0 has been issued at price p0 (b0 ; , B 0 ), the government faces the temptation to
15 The

function p0 (b0 ; , B0 ) is itself defined implicitly by the following equation


p0 = + (1 )( x

Ab0
x +y
+ y)
Ab
B0 p0 0

(16)

This equation has a unique solution (the left-hand side is increasing in p0 while the right-hand side is
decreasing in p0 ), which defines a function p0 (b0 ; , B0 ) which is increasing in b0 , decreasing in B0 and
increasing in .
16 The function p ( b ; , B ) is locally increasing in b if and only if e ( x + y ) < p ( b ; , B ) B . It is easy
0 0
0
0 0
0 0
0
to see that this inequality automatically holds when b0 = A. This implies that it holds for all b0 . Indeed,
suppose that there exists b0 < A such that e( x + y) > p0 (b0 ; , B0 ) B0 . Then as we increase b0 from that
point towards A, p0 (b0 ; , B0 ) keeps decreasing and hence e( x + y) > p0 (b0 ; , B0 ) B0 keeps being verified,
a contradiction. Therefore e( x + y) p0 (b0 ; , B0 ) B0 for all b0 . This in turn implies that p0 (b0 ; , B0 ) is
increasing in b0 .
17 That the function is decreasing in B follows from the fact that the left-hand side of equation (16) is
0
increasing in p0 and independent of B0 , while the right-hand side is decreasing in p0 and decreasing in B0 .
18 That the function is increasing in follows from the fact that the left-hand side of equation (16) is
increasing in p0 and independent of , while the right-hand side is decreasing in p0 and increasing in . To
see that the right-hand side of equation (16) is increasing in , rewrite the right-hand side using equation
(16) as + (1 )( x + y)
19 This

Ab0
p0
Ab0
B0 p
0

B0 ( L)

where B0 ( L) B0 .

is immediate since under commitment and no commitment, all investments are financed, defaults
occur in the same states, and foreigners are as well off. As a result, the sum of consumer welfare and
banking entrepreneur welfare is the same under commitment and no-commitment V0C + V0B = V0C + V0B .
However the welfare of bankers is higher and that of consumers lower under no commitment V0B > V0B
0 = V C + B V B + I A (1 )(1 x y) is greater
and V0C < V0C . Because B < 1, this implies that W
0
0
C
B
B
I

than W0 = V0 + V0 + A (1 )(1 x y).

28

renege by relaxing supervision and lowering the effective diversification requirement b0


below b0 . Foreigners are powerless to resist the re-nationalization of domestic debt unless
they are able to coordinate not to sell their domestic sovereign bonds to domestic banks,
which unlike debt relief negotiations, seems to have few real world counterparts.20
One of the important aspects of banking unions is the transfer of banking supervision from the national to the supranational level. Such a transfer weakens or removes
the temptation of domestic governments to strategically allow their banks to load up
on domestic sovereign bonds to extract larger concessions from legacy creditors. It can
therefore facilitate the implementation of the commitment solution with a high diversification requirement b0 = b0 . This is because the international supervisors objective
function naturally puts more weight on international investors than the domestic government, making it less tempting to relax supervision ex post.21 To make this point starkly,
we study the limit where the supranational supervisor puts full weight on international
investors and no weight on domestic agents. In this limit, the commitment solution is
implemented.
Proposition 9 (Banking Union). If the relaxation of supervision is fully priced in by international investors at the time of the issuance of date-0 debt, then the domestic government faces a
time-inconsistency problem. It is made better off by promising not to engage in supervisory leniency and to set a high effective diversification requirement b0 = b0 before issuing debt at date
0, but it is tempted to relax this requirement after the issuance and lower b0 below b0 . A banking
union removes this temptation and improves welfare (in the illustrating example, welfare increases
0 > W0 ).
from W0 to W

Extensions

The analysis so far has relied on strong assumptions: banks cannot pledge income and
therefore cannot borrow; the government can always finance bailouts; sovereign debt maturity follows an ALM precept of matching maturity and fiscal receipts; foreign investors
are never bailed out; and finally there is a single risky country. This section accordingly
considers a number of extensions of the basic model. We investigate in turn in five extensions of the basic model, the role of leverage, limited bailouts, sovereign debt maturity,
foreign banks in the foreign (safe) country, and multiple risky countries. The first three
20 To the extent that foreign investors are located in different countries, foreign national supervisors would

also need to coordinate in order to facilitate this outcome.


21 Another possibility is that the international supervisor has a better ability to commit to regulation than
the domestic government.

29

extensions do not consider the possibility of debt forgiveness (either because debt is on
the right side of the legacy Laffer curve, or because debt is on the wrong side of the legacy
Laffer curve but investors have difficulties coordinating on a debt relief package). Consequently, there will be no supervisory leniency in Sections 5.1 through 5.3.

5.1

The Role of Leverage

In this section, we introduce leverage into the model. We assume that a fraction 0 (s) I (s)
of the return 1 (s) I (s) is pledgeable to outside international investors at date 1. Banking entrepreneurs can now raise 0 (s) < 1 units of funds per unit of investment at date
1. Consistent with our previous assumptions, we assume that financial claims on 0 (s)
are issued abroad. This can be accommodated by our formalization along the lines of
Holmstrm-Tirole (1997) (see Appendix A). Because we assume away the possibility of
debt forgiveness, there is no supervisory leniency. We make use throughout of the fact
that b0 = b0 .
Leverage and financial shocks. Because banking entrepreneurs can lever up, they only
need a net worth of I (s)(1 0 (s)) in order to invest I (s). As a result, the required bailout
is now
X (s) = max{ I (s)(1 0 (s)) (b0 + p1 (s)b0 ), 0}.
The pricing equation (2) is unchanged, leading to the following fixed-point for the date-1
price p1 (s) of government bonds
p1 (s) = 1 F ( B1 (s)|s),
where
B1 (s) = B0 + max{

I (s)(1 0 (s)) b0
A b0

, 0}.
p1 ( s )
p0

Proposition 10 (Feedback Loop and Leverage). When a fraction 0 (s) of the date-2 return of
the investment project of banking entrepreneurs is pledgeable, the sensitivity of date-1 bond prices
p1 (s) to the state s when a bailout is required is given by

dp1 (s)
=
ds

F ( B1 (s)|s)
s

d[(10 (s)) I (s)]


1
f ( B1 (s)|s)
ds
p1 ( s )
.
I (s)(10 (s))b0
f
(
B
(
s
)|
s
)
1
2
p1 ( s )

Proposition 10 extends Proposition 1 to the case where leverage is positive. The main
difference is that the financing needs I (s) are replaced by I (s)(1 0 (s)). This is simply
30

because banking entrepreneurs can leverage every unit of bailout with private funds by
borrowing 0 (s) units of funds from international investors.
Joint defaults. So far we have ignored the possibility that private debt contracts of banking entrepreneurs might be defaulted upon. In other words, we have assumed that the
enforcement of private debt contracts is perfect. In reality, whether or not to enforce
private contracts is to a large extent a decision by the domestic government. And the
decisions to enforce private debt contracts and to repay sovereign debt tend to be correlated. After all, not enforcing private debt contracts is another way for the government
to default on the countrys obligations.22 We capture this idea by assuming that the costs
of not enforcing debt contracts and to default on sovereign debt take the form of a single
fixed cost. This feature builds in a complementarity between the two decisions. As a result, sovereign defaults come together with defaults on the private debt contracts issues
by banking entrepreneurs, resulting in a positive correlation between bank and sovereign
spreads.
Private debt contracts are priced fairly and reflect the probability that they will not be
enforced. As a result, leverage becomes endogenous. Entrepreneurs can raise 0 (s) p1 (s)
units of funds per unit of investment. The fact that the debt that they raise bears enforcement risk limits their ability to raise funds at date 1, and increases the size of the required
bailout to
X (s) = max{ I (s)(1 0 (s) p1 (s)) (b0 + p1 (s)b0 ), 0}.
The pricing equation (2) is unchanged, leading to the following fixed-point for the date-1
price p1 (s) of government bonds
p1 (s) = 1 F ( B1 (s)|s),
where
B1 (s) = B0 + max{

A b0
I (s)(1 0 (s) p1 (s)) b0

, 0}.
p1 ( s )
p0

Proposition 11 (Feedback Loop and Joint Defaults). When a fraction 0 (s) of the date-2
return of the investment project of banking entrepreneurs is pledgeable and private debt contracts
22 In our model, private financial contracts are between domestic agents (banking entrepreneurs) and foreign agents (international investors). A more general model would also feature private financial contracts
between domestic agents. To the extent that enforcement decisions cannot discriminate between contracts
based on the identities of the parties to the contract, this introduces potential additional costs to the decision
of not enforcing private contracts. These costs are both ex-post in the form of undesirable redistribution and
ex-ante in the form of a reduction in private trade between domestic agents (see e.g. Broner and Ventura
2011). We purposefully stay away from these fascinating issues, which are not the focus of this paper.

31

are defaulted upon when there is a sovereign default, the sensitivity of date-1 bond prices p1 (s) to
the state s when a bailout is required is given by

dp1 (s)
=
ds

F ( B1 (s)|s)
s

d[(10 (s) p1 (s)) I (s)]


1
f ( B1 (s)|s)
ds
p1 ( s )
.
I b0
1 p2 (s) f ( B1 (s)|s)
1

There are two key differences between Proposition 11 and Proposition 10. The first difd[(10 (s) p1 (s)) I (s)]
ference is that the second term in the numerator is now p 1(s) f ( B1 (s)|s)
ds

p 1(s)
1

d[(10 (s)) I (s)]


f ( B1 (s)|s)
,
ds

instead of
reflecting the dependence of the liquidity needs
of banking entrepreneurs on p1 (s) through the pledgeability of returns and leverage. The
second difference is in the denominator. For given values of the date-1 bond price p1 (s), of
the reinvestment need I (s), of the bailout X (s), and hence of date-1 debt B1 (s), the denomI (s)(10 (s) p1 (s))b0
I (s)b
f ( B1 (s)|s).
inator is now smaller at 1 p2 (s) 0 f ( B1 (s)|s) instead of 1
p2 ( s )
1

dp1 (s)
ds

As a result the sensitivity


of the price p1 (s) to the state s is larger.
The feedback loop is stronger, because of a new mechanism operating through the
endogenous leverage of banks. As sovereign risk rises, banks have to reduce leverage.
This is because banks borrowing spreads increase, reflecting the increased probability of
a default on the private debt that they issue. This requires a larger bailout, which puts
further pressure on the government budget etc., ad infinitum.

5.2

Limited Bailouts and Endogenous Diversification

So far, we have maintained the assumption that no matter what portfolios banks hold,
the government can always raise enough funds at date 1 to bail them out completely. We
now relax this assumption.23 We show that when the governments ability to bail out
the banking system is limited, banks naturally limit their exposure to domestic sovereign
default risk.
We assume that there is no debt forgiveness. It is easy to see that the absence of supervisory leniency extends to the setting of this section. We therefore make use of the fact
that b0 = b0 throughout.
To simplify, we assume that I (s) = I is independent of s so that there are no financial
if banks choose b = I,
they do not need
shocks but only fiscal shocks. Because A I,
0
a bailout. But we assume that there are some states of the world where the government
23 Here

it is important to restrict the concept of banking union to the notion of shared supervision;
for, by providing a larger pool of bailout funds a banking union could increase incentives for risk shifting
if the domestic government cannot raise enough funds itself.

32

is not able to fully bail out banks if they choose b0 = b0 . In states of the world s where
funds are insufficient to bail out all the banks, the government optimally bails out as
many banks as possible, saving first the banks with the highest pre-bailout net worth.
This pecking order maximizes the number of banks that can be saved and hence ex-post
welfare.
While banks are ex-ante identical, equilibria can be asymmetric. We therefore look
for an equilibrium in which banking entrepreneurs invest different amounts in foreign
bonds, according to a probability distribution with G with support contained in [b0 , I].
This probability distribution G is an endogenous object, to be solved for as part of the
equilibrium. It might be a degenerate atom, in which case the equilibrium is symmetric.
In every state s, there is an endogenous threshold b0 (s) such that banking entrepreneurs
with b0 b0 (s) secure enough post-bailout funds to finance their investment. This
threshold is monotonically decreasing in s. There is also an endogenous threshold b 0 (s)
b0 (s) such that banking entrepreneurs with b0 b 0 (s) can finance their investment withp (s)
out any bailout. This threshold is defined by I b 0 (s) ( A b 0 (s)) 1p0 = 0, and is also
monotonically decreasing in s.
In states where b0 (s) > b0 so that bailouts are partial, the following bailout equations
must hold

p1 ( s )
1
[ B1 (s) B0 ],
(17)
( A b0 ) dG (b0 ) =
p0
f ( B1 (s)|s)
b0 [b0 (s),b 0 (s))

b0 [b0 (s),b 0 (s))

[ I b0 ( A b0 )

p1 ( s )
]dG (b0 ) = [ B1 (s) B0 ] p1 (s),
p0

where
B1 (s) = B0 +

b0 [b0 (s),b 0 (s))

max{

I b0
A b0

, 0}dG (b0 ).
p1 ( s )
p0

(18)

This simply guarantees that the government determines how much debt to issue at date 1
in order to maximize the number of banks that can be saved.24 Note that the government
necessarily issues less debt than the amount that would maximize the revenues from
this issuance. This is because at the peak of the issuance Laffer curve (the value of B1 (s)
24 Indeed

equation (17) is the first-order condition for the following planning problem:
b0 (s) =

s.t.

b0 b 0 (s)

max{ I b0 ( A b0 )

min

{b 0 (s),B1 (s)}

b 0 (s)

1 F ( B1 (s)|s)
, 0}dG (b0 ) = [ B1 (s) B0 ][1 F ( B1 (s)|s)].
p0

33

which maximizes [ B1 (s) B0 ][1 F ( B1 (s)|s)], a marginal reduction in issuance B1 (s) B0


brings about a second-order reduction in issuance revenues [ B1 (s) B0 ][1 F ( B1 (s)|s)]
1 F ( B1 (s)|s)
but a first-order improvement in banks pre-bailout net worth b0 + ( A b0 )
,
p0
and hence a first-order reduction in required bailouts and by implication a first-order
increase in the number of banks that can be saved.
In addition, the following pricing equations must hold
p1 (s) = 1 F ( B1 (s)|s),

(19)

p0 =

p1 (s)d (s).

(20)

An individual banking entrepreneur who invests b0 gets a bailout in states s > s(b0 )
but no bailout in states s < s(b0 ), where s(b0 ) is the inverse of b0 (s) and is hence monotonically decreasing in b0 . There is another threshold s(b0 ) such that the entrepreneur
does not need a bailout to finance his investment when s > s(b0 ), where s(b0 ) is the inverse of b 0 (s) and is hence monotonically decreasing in b0 . This banking entrepreneur
now faces a meaningful tradeoff in his portfolio decision. By increasing his investment b0
in foreign bonds, he secures a bailout in some states of the world where he did not get a
bailout by rising in the government bailout pecking order, but loses out in states where he
does not need a bailout to fund his investment. The corresponding optimality conditions
states that b0 maximizes his welfare25
b0 arg max V0B (b0 (i )),
b0 (i )

25 For b
0

in the interior of the support of G, must be absolutely continuous with respect to the Lebesgue
measure in the neighborhood of s(b0 ) with Radon-Nikodym derivative d (s) = (s)ds, and the entrepreneur must be left indifferent by marginal changes in b0 , which requires that the following differential
equation in s(b0 ) hold on the interior of the support of G:


 

p1 (s(b0 ))
p1 ( s )
s0 (b0 ) (s(b0 )) 1 (s(b0 )) I b0 + ( A b0 )
=
1 (s)ds.
p0
p0
s(b0 )
The left-hand-side represents the marginal utility gain from securing bailouts in more states of the world,
while the right-hand-side represents the utility loss in states where no bailout is required to fund the investment.

34

where

V0B (b0 (i ))

(s) +
1 (s) Id

{ss(b0 (i ))}

[b0 (i ) + ( A b0 (i ))

{s<s(b0 (i ))}

p1 ( s )
I ]d (s)
p0

[b0 (i ) + ( A b0 (i ))

p1 ( s )
1 (s) I]d (s).
p0

The determination of equilibrium resembles that of equilibria of full-information firstprice auctions or wars of attrition. The complication here comes from the fact that the
object that competitors vie forhere subsidiesis itself endogenous, as from equation
(17), the pot of subsidies depends on the distribution of bids, namely the holdings of
foreign bonds.
An interesting feature of these equilibria is that they display a force for endogenous
diversification. Banking entrepreneurs choose to hold foreign bonds even in the absence
of regulation. This is because they cannot be certain to count on a government bailout. We
illustrate this possibility with two simple examples. In the first example, the distribution
G is a degenerate atom. In the second example, it is non-degenerate. In both cases, we
abstract away from regulation and set b0 = 0.26
Illustrating example 1. Our first example is a variant of the example in Section 2.1. We
assume that ( E B0 ) x < ( x + y)(e B0 ), so that the revenue maximizing level of B1 ( L)
in state L is e.
Our candidate equilibrium is symmetric with B1 ( L) = e, p1 ( L) = x + y and p0 =
+ (1 )( x + y).27 The limited-bailout condition is
(1 x y )
( A b0 ) = (e B0 )( x + y).
+ (1 )( x + y)
26 Another

(21)

form of bailout rat race is developed in Nosal-Ordonez (2014). In their paper as in ours, the
government ex ante dislikes bailing out banks, but cannot help doing so when faced with the fait accompli.
The innovation of their paper is that a bank can be rescued either by the government or (more cheaply) by a
healthy bank and the government prefers a private takeover to a public takeover. The government however
does not know whether the first distressed banks shock is idiosyncratic or aggregate (in which case there
will be no healthy bank to rescue the distressed one). In a situation in which the conditional probability
of an aggregate shock is not too large, the government waits, and therefore banks prefer not to be the first
distressed institution. If they can sink resources to augment the probability of not being first, they will do
so, a behavior akin to a rat race.
27 It can be shown that there are no asymmetric equilibria in this example.

35

In order for bankers to prefer b0 to 0, we must have


[

1
x+y
1]b0 (1 ) A[1
].
+ (1 )( x + y)
+ (1 )( x + y)

(22)

The solution b0 of equation (21) always (strictly) verifies equation (22). This guarantees
that our candidate equilibrium is indeed an equilibrium as long as the solution of equation (21) verifies 0 < b0 < A.
Illustrating example 2. We now consider a simple variant of the previous example. The
structure of uncertainty is as follows. With probability , the state s is H and the endowment is high enough at E that there is no default. With probability (1 )z, the state is M,
and the endowment is high enough at E so that there is no default with conditional probability x, intermediate e M with conditional probability y, and 0 with conditional probability 1 x y. With probability (1 )(1 z), the state is L, and the endowment is
high enough at E so that there is no default with conditional probability x, intermediate
e L with conditional probability y, and 0 with conditional probability 1 x y. What distinguishes states M and L is that e M > e L . We assume that ( x + y)(e L B0 ) > ( E B0 ) x
so that the revenue maximizing level of debt is e M in state M and e L in state L.
Our candidate asymmetric equilibrium is such that there are full bailouts in the medium
state, but limited bailouts in the low state. Bankers invest b 0 ( L) with probability and 0
with probability 1 . Prices are p0 = + (1 )( x + y), p1 ( L) = p1 ( M) = x + y.
The bailout conditions are

(1 x y )
( A b 0 ( L)) = (e L B0 )( x + y),
+ (1 )( x + y)

(23)

(1 x y )
(1 x y )
( A b 0 ( L)) + (1 )
A (e M B0 )( x + y).
+ (1 )( x + y)
+ (1 )( x + y)
(24)
In order for bankers to be indifferent between b0 = b 0 ( L) and b0 = 0, we must have

1
x+y
1]b 0 ( L) = A(1 z)(1 )[1
].
+ (1 )( x + y)
+ (1 )( x + y)

We can rewrite equation (25) as


+ (1 )( x + y)
b 0 ( L) = A(1 z)[(1 1)
+ 1].
(1 x y )

36

(25)

Using equation (23), we find


+(1 )( x +y)

( x + y ) (1 x y )
e L B0
=
.
A 1 (1 z)[(1 1) +(1 )( x+y) + 1]
(1 x y )
We have an equilibrium if b 0 ( L) < A, 0 < < 1, and

(1 )

(1 x y )
A (e M e L )( x + y),
+ (1 )( x + y)

which can always be ensured for appropriate parameter values.


Proposition 12 (Bailout Rat-Race and Incentives for Diversification). In the illustrating
examples with limited bailouts and symmetric or asymmetric equilibria, it is optimal for banks
to not fully load up on domestic sovereign default risk and instead choose a non-zero degree of
diversification b0 > 0 with positive probability even when there is no regulation (b0 = 0).

5.3

Sovereign Debt Maturity

In this section, we investigate the role of sovereign debt maturity. More specifically, we
compare our economy with long-term sovereign bonds which are claims to coupons accruing at date 2 with an economy where sovereign bonds are short-term one-period bonds
which are rolled over at date 1.
We assume that there is no debt forgiveness. It is easy to see that the absence of supervisory leniency extends to the setting of this section. We therefore make use of the fact
that b0 = b0 throughout.
With long-term bonds, welfare is given by
"
W0 =

B1 (s)

[ E B0 ] f ( E|s)dE +

B1 (s)

#
I

[ E ] f ( E|s)dE + (s) I (s) d (s)

B [1 (s) I (s) + [ A I (s)]] d (s)

p (s)
+ (1 B ) min{b0 + ( A b0 ) 1
I (s), 0}d (s).
p0

We now consider the economy with short-term bonds. We denote all variables with a
tilde. To make the comparison with the economy with short-term bonds meaningful, we

37

impose that the government must raise the same amount of revenues G0 in period 0, i.e.

B 0 = B0

[1 F ( B1 (s)|s)] = G0 .

(26)

In addition, the government must raise exactly enough revenues at date 1 to repay the
date-0 debt that is coming due, i.e. we must have for all s28
B 1 (s)[1 F ( B 1 (s)|s)] = B 0 .

(27)

Because I A, no bailouts are required. Welfare is given by


"
0 =
W

B 1 (s)

E B 1 (s) f ( E|s)dE +

B 1 (s)

#
I

[ E ] f ( E|s)dE + (s) I (s) d (s)

B [1 (s) I (s) + [ A I (s)]] d (s).

Using (26) and (27), we can write

0 =
W0 W

[ F ( B 1 (s)|s) F ( B1 (s)|s)]d (s)

p (s)
I (s), 0}d (s). (28)
+ (1 B ) min{b0 + ( A b0 ) 1
p0

There are two terms on the right-hand side of equation (28). The first term represents the
difference in default costs, and the second term represents the welfare impact of transfers from consumers to banking entrepreneurs resulting from bailouts when domestic
sovereign bonds are long term. In the proof of the proposition below, we show that under
some additional assumptions on the distributions of E and s, the first term is positive.29
The second term is always negative and arises because by issuing short-term bonds, the
government reduces the risk-taking possibilities of banks and insulates the banks from
fiscal developmentsthere is no feedback loop between banks and the sovereign. As a
result, the intuitive asset-liability management (ALM) principle of matching maturities of
incomes and payments holds when the minimum diversification requirement b0 is high
28 To

give short-term debt a good shot, we assume that the government is always able to roll over its
short-term debt. This is indeed the case if negative shocks s are not too catastrophic, so that the debt can be
rolled over by pledging income in the good realizations at date 2.
29 The additional assumptions are that F ( E | s ) = F ( E s ) where F is increasing and convex.
( f ( E|s)/(1 F ( E|s)))
These assumptions, which imply the monotone hazard rate properties
0 and
s
( f ( E|s)/(1 F ( E|s)))
E

> 0, are sufficient but not necessary to prove the result.

38

enough: Long-term debt then leads to a strictly lower expected probability of default than
short-term debt.
Intuitively, a short maturity has both benefits and costs. The cost is that a short maturity is bad for fiscal hedging. The benefit is that a short maturity reduces the risk-shifting
possibilities of banks. A short maturity is therefore a costly substitute to supervision. As
a result, a long maturity is preferable when supervision is efficient enough (b0 is high
enough).
Proposition 13 (Optimal Debt Maturity). Suppose that F ( E|s) = F ( E s) where F is increasing and convex. Then for b0 high enough, welfare is higher with long-term sovereign bonds
0.
than with short-term sovereign bonds W0 > W
while
Proof. Note that B1 (s) B0 and that B1 (s) converges to B0 for all s as b0 goes to I,

B 1 (s) is independent of b0 . Hence the result follows if we can show that [ F ( B 1 (s)|s)
F ( B0 |s)]d (s) < 0. We now proceed to prove this result, which we refer to as result A.
The result is a direct consequence of the following related (dual) result, which we refer to
as result B. Let B0 be defined by

B0 [1 F ( B0 |s)]d (s) = G0
as above, and let B 1 (s) and G 0 be defined by the system of equations
B 1 (s)[1 F ( B 1 (s)|s)] = G 0

F ( B 1 (s)|s)d (s) =

for all s,

F ( B0 |s)d (s).

Result B is that G 0 < G0 . We now prove result B, which in turn directly implies result A.
Since

B0

d (s) = G0 ,
G0

B1 (s)
we need to show that

B0
B 1 (s)

d (s) > 1.

By Jensens inequality, result B is implied by the following result, which we refer to as


result C:
B0
> 1.

B1 (s)d (s)

39

Since
to

F ( B 1 (s) s)d (s) = F ( B0 s)d (s) and F is increasing, result C is equivalent

B 1 (s0 )d (s0 ) s)d (s) <

F(

F ( B 1 (s) s)d (s).

Define

g() =

F[

B 1 (s0 )d (s0 ) + ( B 1 (s)

B 1 (s0 )d (s0 )) s]d (s).

We have


0
0
0
0
0

g () =
f [ B1 (s )d (s ) + ( B1 (s) B1 (s )d (s )) s][ B1 (s) B 1 (s0 )d (s0 )]d (s).

Because f [ B 1 (s0 )d (s0 ) + ( B 1 (s) B 1 (s0 )d (s0 )) s] and B 1 (s) B 1 (s0 )d (s0 ) are
decreasing in s for all 0, the right-hand side is the covariance of two decreasing
functions of the random variable s and is therefore positive. It follows that g0 () > 0 for

all 0. Since g(0) = F ( B 1 (s0 )d (s0 ) s)d (s) and g(1) = F ( B 1 (s) s)d (s), we
get result C. Results B and A follow, concluding the proof of the Proposition.

There are obvious extensions of our setup that would reinforce the conclusion of
Proposition 13. For example, the desirable features of short-term sovereign debt in terms
of limiting the risk-taking possibilities of banks, bailouts and feedback loops between
banks and sovereigns, would be mitigated in a model with a richer set of risk taking
possibilities apart from domestic sovereign debt, or in an infinite horizon version of our
model with overlapping generations of banking entrepreneurs, consumers and investors,
where some banks hold domestic sovereign debt for liquidity in all periods.

5.4

Foreign Banks in the Foreign (Safe) Country

In our basic model, we abstracted from foreign banks in the foreign (safe) country. We can
introduce such banks. These face a similar problem to domestic banks. They have some
net worth A F at date 0, and some investment opportunities I F (s) at date 1 with private
and foreign social returns given by 1F (s) and I,F (s). We assume that with A F IF where
IF = maxsS I F (s). Foreign banks invest their net worth at date 0 in a portfolio of risky
domestic bonds (bonds of the domestic economy) and safe foreign bonds (bonds of
the foreign economy).30 The return on their portfolio at date 1 determines their net worth
30 Of course domestic bonds are foreign bonds from the perspective of foreign banks, and similarly foreign

bonds are domestic bonds from the perspective of foreign banks. To avoid confusion, we always refer to

40

at date 1. If it falls short of their investment need, then they are bailed out by the foreign
government. But these bailouts do not endanger the ability of the foreign government
to repay its debt. The domestic and foreign countries differ only in the riskiness of their
sovereign bonds. Domestic sovereign bonds are risky and foreign sovereign bonds are
safe. We denote by b0F the supervisory capability of the foreign government, and by b0F
the effective minimum diversification requirement.
Our analysis goes through in this extended model as long as risk-neutral international investors who do not benefit from bailout guarantees (not foreign banks) remain
the marginal buyers of domestic and foreign sovereign bonds. In particular, Propositions 1-13 still hold without any modification. The key observation is that foreign banks
portfolio decisions are irrelevant for equilibrium prices, domestic bailouts and sovereign
default probabilities, and domestic banks portfolio decisions. Foreign banks risk exposures do not give rise to any feedback loop, because the foreign government has enough
fiscal capacity to bail them out without endangering its ability to repay its debt.
The extended model has additional predictions on the incentives of foreign banks and
of the foreign government. Because of the bailout guarantees, foreign banks have an
incentive to load up on risky domestic debt. The foreign government has an incentive to
regulate foreign banks so that they do not take on too much domestic sovereign risk. We
elaborate on these issues now.
Frictionless supervision of foreign banks in the foreign (safe) country. We first consider first optimal supervision in the foreign (safe) country as in Section 3.3. We can derive
the following equivalent of Propositions 2 and 8.
Proposition 14 (Supervision in Foreign (Safe) Country). When the basic model in Section 3.3
is extended to include foreign banks in the foreign (safe) country, there is no supervisory leniency
in the foreign (safe) country so that b0F = b0F . Moreover when the supervisory capability b0F
can be chosen at no cost by the foreign government (first-best frictionless supervision), it is optimal
to set b0F = IF . All these statements are true whether or not debt forgiveness is allowed.
Just like the domestic government in the absence of debt forgiveness, the foreign government has an incentive to prevent its banks from taking on (domestic) sovereign risk.
This is because when foreign banks take on more risk, they receive a bailout from the
foreign government following a bad shock, which has adverse distributional effects.
domestic bonds as the sovereign bonds of the domestic economy, independently of whether they are held
by domestic or foreign agents. Similarly, we refer to foreign bonds as the bonds of the foreign economy,
independently of whether they are held by domestic of foreign agents.

41

In the presence of debt forgiveness, Proposition 8 displays a sharp contrast between


the supervisory incentives of the domestic government and those of the foreign government. Because foreign government debt is safe, the foreign government cannot extract
any concessions from its creditors. As a result, the foreign government has no incentive to
engage in strategic supervisory leniency. Instead it always seeks to strictly limit the exposure of foreign banks to domestic sovereign risk. Specializing the model to the illustrating
example, the implication of the extended model is then that as the probability 1 of a
bad domestic fiscal shock increases, domestic supervision of domestic banks gets laxer,
but foreign supervision of foreign banks does not, and as a result domestic banks tilt their
portfolios towards risky domestic bonds and away from safe foreign bonds, but foreign
banks do not.
Collective moral hazard and foreign banks in the foreign (safe) country. It is also interesting to investigate the portfolio decisions of foreign banks in the environment of Section
3.4, assuming that foreign banks face a cost of making their balance sheets opaque F similar to that of domestic banks and that I F (s) = IF is independent of s and that A F = IF .
Using the fact that b0F = b0F , we can derive the following equivalent of Proposition 5.
Proposition 15 (Multiple Equilibria). When the illustrating example of Section 3.4 is extended
to include foreign banks in the foreign (safe) country, the portfolio of foreign banks is given by
(1 )(1 )
b0F = b0F ( F0 )1 ( +(1 )( ) ) with = x in the low (domestic) diversification equilibrium
and = x + y in the high (domestic) diversification equilibrium.
Foreign banks exposure to domestic sovereign risk is higher in the low (domestic)
diversification equilibrium than in the high (domestic) diversification equilibrium.31
The key observation that underlies these results is that there are strategic complementarities running from domestic banks to foreign banks, but no strategic complementarities running in the other direction. Indeed, when domestic banks increase their exposure
to domestic sovereign risk, the benefits of doing so also increases for foreign banks. But
when foreign banks increase their exposure to domestic sovereign risk, the benefits of
doing for domestic banks remains unchanged. This is because the riskiness of domestic
debt increases in the former case but not in the latter.
This also implies that there are supervisory externalities running from the domestic
country to the foreign (safe) country but not vice versa. Indeed, suppose that at some
supervisory cost R (respectively R F ), the domestic (respectively foreign) government can
31 Note

that contrary to domestic and foreign banks, international investors have less exposure to domestic sovereign risk in the low (domestic) diversification equilibrium than in the high (domestic) diversification equilibrium.

42

achieve perfect supervision with supervisory capability b0 = A (respectively b0 F = A F ),


in which case, because our example assumes that there is no debt forgiveness, we also
have b0 = b0 = A (respectively b0 F = b0 F = A F ). Otherwise, supervision is inexistent
( and F are both zero), so that banks can perfectly evade regulation. Assume that
(1 x )
B0 + 1x +(1 ) x A > e > B0 .
If the domestic government chooses to incur the supervisory cost R, we have B1 ( L) =
(1 x )
B0 and = x + y. Otherwise B1 ( L) = B0 + 1x +(1 ) x A and = x. In both cases, we have
p1 ( L) = and p0 = + (1 ).
(1 )
The net gain (1 )(1 B ) A F +(1 ) R F from incurring the supervisory cost for
the foreign government is lower ( = x) when the domestic government incurs the supervisory cost than when it doesnt ( = x + y). By contrast, the net gain from incurring the
supervisory cost for the domestic government is independent of whether or not the domestic government incurs the supervisory cost. More interestingly, we have the following
proposition.
Proposition 16 (Supervisory Externalities and Banking Union). In the illustrating example with either perfect or irrelevant supervision and foreign banks in the foreign (safe) country,
foreign welfare increases with the supervisory effort (decreases with the supervisory cost) of the
domestic country, but domestic welfare is independent of the supervisory effort (independent of the
supervisory cost) of the foreign country.
Proposition 16 uncovers an additional rationale for a banking union. Domestic supervision has positive external effects for the foreign country. These effects are not internalized by the domestic government, and as a result, supervision is too lax in the domestic
economy. By transferring supervisory decisions from the national to the international
level, a banking union allows these effects to be internalized, leading to a toughening of
supervision in the domestic country and an improvement of welfare.

5.5

Multiple Risky countries

As discussed in the introduction, risk shifting by a troubled European countrys banks


could have occurred through the purchase of another troubled countrys bonds rather
than through a re-nationalization of the domestic financial market.
In this section, we therefore considers multiple risky countries. We show that, provided that balance sheet shocks and fiscal shocks within a country are at least slightly positively correlated and that fiscal shocks across countries are not perfectly correlated (a reasonable assumption), risk shifting solely through domestic bond holding is a strict equilibrium. We also show that, with multiple risky countries, our double-decker bailout
43

theory predicts that when the fiscal outlook is bad, governments in risky countries have
an incentive to relax supervision and let their banks load up on risky domestic debt (and
not risky foreign debt).
All in all, this extension shows that the multiple forces that we have identified for risk
shifting in the baseline model occur through the purchase of risky domestic debt rather
than risky foreign debt, implying that the re-nationalization results of the baseline model
are robust to multiple risky countries.
The structure of the model is the same as in the basic model, but there are now two
symmetric risky countries A and B, together with the foreign (safe) country. We consider
banks in countries A and B, but for simplicity, we abstract from banks in the foreign (safe)
country. We denote by si the state of the world at date 1 in country i, and we denote by
the joint distribution of (s A , s B ). We focus on symmetric equilibria throughout.
In each country i { A, B}, banks have an endowment A at date 0 and some investment opportunities I (si ) with private and social returns (for country i) given by 1 (si ) and
I (si ). Banks invest their net worth at date 0 in a portfolio of safe foreign bonds, risky domestic bonds and risky foreign bonds. The return on their portfolio at date 1 determines
their net worth at date 1. If it falls short of their investment need, then they are bailed out
by their countrys government.
We denote by B0 the quantity of debt at date 0 and by p0 the price of debt at date 0
in both risky countries. We assume that in each country i { A, B}, supervision forces
banks of country i to hold a portfolio with holdings of safe foreign sovereign bonds of at
least b0 b0 , but does not place constraints on the relative holding of risky sovereign
debt of countries A and B. We start by assuming that there is no debt forgiveness (we
reintroduce this possibility later). Each government chooses not to engage in supervisory
leniency and sets b0 = b0 , which we assume from now on.
f
In each country i, we denote by b0 b0 , b0d 0 and b0 0 the holdings of foreign
safe debt, risky domestic debt (of country i) and risky foreign debt (of country i) with
f
b0 + p0 (b0d + b0 ) = A. We denote by B1 (sd , s f ) the quantity of debt at date 1 and by
p1 (sd , s f ) the price of debt at date 1 when the domestic when country i is in state si = sd
and country i is in state si = s f .
A symmetric equilibrium is characterized by the following pricing equations
p1 (sd , s f ) = 1 F ( B1 (sd , s f )|sd ),

p0 =

p1 (sd , s f )d (sd , s f ),

44

with

b0 p1 (s f , sd )
b0d
I (sd ) b0
B1 (s , s ) = B0 + max{

, 0},

p0
p0 p1 ( s d , s f )
p1 ( s d , s f )
d

together with the requirement that bank portfolios (b0 , b0d , b0 ) solve the following maximization problem
max

h
f

(b 0 ,b 0d ,b 0 )

i
f
1 (sd ) I (sd ) + max{b 0 + p1 (sd , s f )b 0d + p1 (s f , sd )b 0 I (sd ), 0} d (sd , s f ),

subject to
f
b 0 + p0 (b 0d + b 0 ) = A

and

b 0 b0 .

Illustrating example with multiple risky countries. For simplicity, we carry out our
multiple-country extension in the context of our illustrating example. There are two states
H and L at date 1 for each country with probability and 1 . Let k H be the probability
that country B is in state H if country A is in state H. Let k L be the probability that country
B is in state L if country A is in state L. Symmetry imposes that

(1 )(1 k L ) = (1 k H ).
We assume that k H < 1 and k L < 1 so that shocks in the two countries are not perfectly
correlated.
In state H in country i { A, B}, date-2 fiscal revenues are equal to E with probability
1, and investment needs are equal to I and the return on investment is equal to 1H . In
state L in country i { A, B}, date-2 fiscal revenues are equal to E with probability x,

e with probability y, and 0 with probability 1 x y, investment needs are equal to I,


and the return on investment is equal to 1H . We assume that I < I = A. Hence fiscal
and balance sheet shocks are positively correlated in a given country. As will be clear
but the size of the gap between I and I
below, it is important for our results that I < I,
is not important. In other words, it only matters that there be some positive correlation
between balance sheet and fiscal shocks. Although this is not important, we assume that
1H I > 1L I so that state H (respectively L) corresponds to a state with high (respectively
low) future profits but low (respectively high) liquidity needs.
Home bias with multiple risky countries. We assume that E is large enough so that
p1 ( H, H ) = 1 and p1 ( H, L) = 1. But we have p1 ( L, L) < 1 and p1 ( L, H ) < 1. We show
that as long as p1 ( L, H ) is not too high, then banks in country j choose to hold as little
45

safe foreign bonds and as much risky domestic bonds as allowed by supervision, but no
risky foreign bonds.
Proposition 17 (Home Bias with Multiple Risky Countries). Consider the illustrating example with two symmetric risky countries i { A, B}. Then there exists a symmetric equilibrium in
which p1 ( L, H ) p0 and banks in each country i { A, B} choose to hold as little safe foreign
bonds and as much risky domestic bonds as allowed by supervision, but no risky foreign bonds:
f
Ab
b0 = b0 , b0d = p0 0 and b0 = 0. This equilibrium is strict. Moreover, there are no other
symmetric equilibria with p1 ( L, H ) p0 .
Proof. We show that in any symmetric equilibrium as long as p1 ( L, H ) p0 , banks in
Ab
each country i { A, B} prefer to choose the following portfolio: b0 = b0 , b0d = p0 0
f

and b0 = 0. Together with the fact that when banks do indeed choose this portfolio,
p1 ( L, H ) = p1 ( L, L) < p0 < p1 ( L, H ) = p1 ( H, H ) = 1, this proves the proposition.
Consider a country i { A, B}. For the same reasons as in the main model, banks
in country i will choose holdings of safe sovereign bonds of exactly b0 . The payoff of a
f
f
banking entrepreneur in country i from holding portfolio (b0d , b0 ) with b0d + b0 = b0 where
Ab
b0 = p0 0 is
f

k H [1H I + max{ A I + (1 p0 )b0d + (1 p0 )b0 , 0}]


f
+ (1 )k L [1L I + max{ A I + ( p1 ( L, L) p0 )b0d + ( p1 ( L, L) p0 )b0 , 0}]
f

+ (1 k H )[1H I + max{ A I + (1 p0 )b0d + ( p1 ( L, H ) p0 )b0 , 0}]


f
+ (1 )(1 k L )[1L I + max{ A I + ( p1 ( L, H ) p0 )b0d + (1 p0 )b0 , 0}].

Only the last two terms of the expression above matters for portfolio choice of the banking
f
entrepreneur. The sum of the last two terms is a convex function of b0d and b0 . The optimal
portfolio is therefore necessarily a corner solution (b0 , 0) or (0, b0 ). We now compute the
value of the sum of the last two terms at these two corners.
For b0d = b0 , the value of the sum of the last two terms is
(1 k H )[1H I + max{ A I + (1 p0 )b0 , 0}]

+ (1 )(1 k L )[1L I + max{ A I + ( p1 ( L, H ) p0 )b0 , 0}].


this can be re-expressed as
Since A I + ( p1 ( L, H ) p0 )b0 0 (recall that A = I),

(1 k H )[1H I + A I + (1 p0 )b0 ] + (1 )(1 k L )1L I.


46

For b0d = 0, the value of the sum of the last two terms is
(1 k H )[1H I + max{ A I + ( p1 ( L, H ) p0 )b0 , 0}]

+ (1 )(1 k L )[1L I + max{ A I + (1 p0 )b0 , 0}].


If A I + ( p1 ( L, H ) p0 )b0 0, this can be re-expressed as
(1 k H )1H I + (1 )(1 k L )[1L I + A I + (1 p0 )b0 ],
which is less than the value with b0d = b0 . If A I + ( p1 ( L, H ) p0 )b0 > 0, this can be
re-expressed as
(1 k H )[1H I + A I + ( p1 ( L, H ) p0 )b0 ] + (1 )(1 k L )[1L I + A I + (1 p0 )b0 ]

= (1 k H )[1H I + A I + (1 p0 )b0 ] + (1 )(1 k L )[1L I + A I + ( p1 ( L, H ) p0 )b0 ],


which is again less than the value for b0d = b0 .
f
Therefore the banking entrepreneur in country j chooses b0d = b0 and b0 = 0.
The intuition for Proposition 17 is simple. Because balance sheet shocks in a given
country are perfectly correlated with fiscal shocks in this country, but imperfectly correlated with fiscal shocks in other countries, banking entrepreneurs maximize the bailout
that they extract from the government by investing in risky domestic sovereign bonds
rather than risky foreign sovereign bonds.32
In this setting, because as shown in Proposition 17, banks in country i { A, B} choose
to hold no debt from country i, there is no interaction between the two risky countries
A and B: Debt prices and quantities in each risky country are determined independently
exactly as in the illustrating example in Section 2. All of our results in Section 3, properly
specialized to the setting of this illustrating example, carry through with no modification.33 The results in Section 5.4 also readily extend. Our other extensions in Sections
5.1-5.3 would require further adaptation.
The results in Section 3.4 are particularly interesting in this context. Propositions 5
and 6 apply with no modification. Switching from the high to the low diversification
32 The

condition that p1 ( L, H ) p0 is equivalent to the assumption that there are bailouts when state
L occurs in country i and state H occurs in country i if banks of country i choose portfolio b0 = b0 ,
Ab

b0d = p0 0 and b0 = 0. Proposition 17 then shows that there exists a unique symmetric equilibrium that
satisfies this condition, and that in this equilibrium, banks choose the aforementioned portfolio.
33 Some of our results were proved in the case where I ( s ) is independent of s whereas we require here
In these cases, we can simply take the limit when I goes to I from below.
that I < I.

47

equilibrium involves a re-nationalization of sovereign debt markets: Banks in country


i increase their holdings of risky domestic sovereign debt from country i but not their
holdings of risky foreign sovereign debt from country i (which are equal to zero in both
equilibria).
Strategic supervisory leniency. We now introduce the possibility of debt forgiveness
and examine how the results in Sections 4.2 and 4.3 regarding strategic supervisory leniency generalize to a setting with multiple risky countries. In particular, we want to
show that governments have an incentive to let their own banks load up on domestic
risky bonds but not on foreign risky bonds. To make that point in the starkest possible
way, we assume that supervision is perfect: in each country i, the government can exf
actly control the portfolio (b0 , b0d , b0 ) of its banks through supervision. This means that
we not only assume that supervisory capability as we have defined it so far is perfect so
that b0 = A, but also in addition that the government can now perfectly determine the
relative holding of risky sovereign debt of countries A and B.
y
We assume that e(1 + x ) > B0 > e and that I (s) = I is independent of s. For
simplicity, we consider the limit where I = I = A. We can derive the following counterpart to Proposition 8, which shows that governments in risky countries have an incentive
to let their banks load up on risky domestic sovereign debt as opposed to risky foreign
sovereign debt, in order to maximize the concessions from legacy creditors.
Proposition 18 (Strategic Supervisory Leniency with Multiple Risky Countries). Consider
the illustrating example with two risky countries and assume that B0 < E. In the limit where A
is small compared to e and B0 , it is optimal for the government in country i to force its banks to
invest all their net worth A in risky domestic sovereign bonds, and to invest zero in safe foreign
f
sovereign bonds and zero in risky foreign sovereign bonds: b0 = 0, b0d = pA0 and b0 = 0.
Proof. We consider a symmetric equilibrium, and denote with a tilde the equilibrium valf
ues, assuming, as we will verify below, that b 0 = 0. And we look at the incentives of the
government in country i { A, B} to deviate from this equilibrium. Banks in country i
f
f
f
invest in portfolio (b0d , b0 , b0 ) with b0d + b0 = b0 and p0 b0d + p 0 b0 + b0 = A. The values of
f
b0d , b0 and b0 are controlled by the government in country i.
We assume that E is large enough that the price of debt in a given country is always
one when this country is in state H (this condition is guaranteed to hold in the limit where
A is small compared to e and B0 ).
If we have debt forgiveness in state ( L, H ), then the post-debt forgiveness amount of

48

debt B0 ( L, H ) satisfies
B0 ( L, H ) +

p0 ( x + y )
x+y

B0 ( L,H )
B0
b0d

p 0 1 f
b = e,
x+y 0

and B1 ( L, H ) = e. If we have debt forgiveness in state ( L, L), then the post-debt forgiveness amount of debt B0 ( L, L) satisfies

B0 ( L, L) +

p0 ( x + y )
x+y

B0 ( L,L)
B0
b0d

B ( L,L)
p 0 ( x + y) 0 B0

x+y

b0 = e,

and B1 ( L, L) = e.
There is debt forgiveness provided that the following conditions are verified:

( x + y) B0 ( L, H ) xB0 ,
and

( x + y) B0 ( L, L) xB0 .
These conditions are always verified when A is small enough compared to e and B0 ,
f
It is then always optimal for the government to choose at date 0 the values of (b0d , b0 )
that maximize welfare (taking p0 as given)
W0 = ( E B0 ) + (1 )[ x ( E e) (1 x y)] + I A + B 1 A
f

+ k H G {(1 p0 )b0d + (1 p 0 )b0 }


B ( L, L)
B0 ( L, L)
f
p0 )b0d + (( x + y) 0
p 0 )b0 }
B0
B0
B ( L, H )
f
p 0 )b0 }
+ (1 k H ) G {(1 p0 )b0d + (( x + y) 0
B0
B ( L, H )
f
+ (1 )(1 k L ) H {(( x + y) 0
p0 )b0d + (1 p 0 )b0 },
B0

+ (1 )k L H {(( x + y)

where G ( x ) = B max{ x, 0} + min{ x, 0} and H ( x ) = B max{ x, 0}. This defines two


f
functions b0d ( p0 ) and b0 ( p0 ).
The date-0 price p0 is then given by the fixed-point equation
p0 = + (1 )( x + y)

(1 k L ) B0 ( L, H ) + k L B0 ( L, L)
,
B0

49

with B0 ( L, H ) and B0 ( L, L) defined as above with b0d = b0d ( p0 ) and b0 = b0 ( p0 ).


In the limit where A is small compared to e and B0 , the last four terms of the expression
for welfare W0 can be rewritten (up to a first order approximation),
e
e
f
))b0d + (1 ( + (1 )( x + y) ))b0 }
B0
B0
e
e
e
e
f
+ (1 )k L H {(( x + y) ( + (1 )( x + y) ))b0d + (( x + y) ( + (1 )( x + y) ))b0 }
B0
B0
B0
B0
e
e
e
f
+ (1 k H ) G {(1 ( + (1 )( x + y) ))b0d + (( x + y) ( + (1 )( x + y) ))b0 }
B0
B0
B0
e
e
e
f
+ (1 )(1 k L ) H {(( x + y) ( + (1 )( x + y) ))b0d + (1 ( + (1 )( x + y) ))b0 }.
B0
B0
B0
k H G {(1 ( + (1 )( x + y)

The solution is clearly b0d =

A
+(1 )( x +y) Be

and b0 = 0.

Obviously, if the government could not determine the relative holdings of domestic
and foreign risky bond holdings, but only impose an effective requirement b0 b0 with
b0 b0 , then we would obtain (in the limit where I tends to I = A from below) that it is
optimal to set b0 = 0. Banks would then by themselves load up on domestic risky bonds,
f
choosing b0d = +(1A)( x+y) e and b0 = 0. Proposition 18 shows the more interesting
B0

result that even if the supervisor could perfectly control the portfolios of banks, it would
choose to encourage them to load up on domestic risky bonds.

Summing Up

We built on a relatively standard model of feedback loop, with shocks reducing the value
of sovereign debt, leading to bailouts if bank portfolios exhibit a home bias, leading to
further debt sustainability problems, etc. Relative to the earlier literature, we have uncovered the following insights:
The feedback loop paradoxically stems from a prudent matching of debt maturity
with the countrys fiscal capability.
The feedback loop is stronger in case of joint default on private and sovereign debt.
As long as the country has the capability to bail out banks, the latters exposures to
their domestic governments debt are strategic complements.
When push comes to shove and the government may run out of money to finance
bailouts, banks may by contrast engage in a diversification rat race.
50

There are two distinct rationales for re-nationalization. First, the banks may invest
in opacity and try to evade prudential diversification rules. Second, even when
the government can perfectly monitor its banks, the government may strategically
turn a blind eye to their lack of country diversification and count on legacy debt
forgiveness to finance the rescue of its banking sector in case of difficulties. In either
case, re-nationalization occurs when the legacy debt increases or prospects about
the countrys fiscal capability worsen. Furthermore, re-nationalization is robust to
the distress of multiple countries and therefore to the co-existence of multiple ways
of shifting risk through the holding of sovereign bonds.
There are two distinct rationales for a banking union. First, if the ex-post leniency of
domestic supervisors is anticipated ex ante at the time of sovereign debt issuance,
then it is priced in the form of higher spreads. The government is better off committing ex ante to a tough ex-post supervisory stance, but is tempted to relax it
ex post. If the government lacks commitment, then it benefits from relinquishing
its supervisory powers to a supranational supervisor by joining a banking union.
Second, voluntary or involuntary supervisory leniency makes the supervision of
foreign banks portfolios more complex, generating an externality that can only be
internalized in a banking union.
Mapping the theory with the European experience is complex. The blanket guarantees
granted by the ECB in the middle of the crisis (akin to the debt forgiveness in the model)
have so far prevented it from unfolding. Nonetheless the existing empirical evidence is
encouraging. Acharya et al. (2013) looks at the price of European sovereign and bank
CDSs over the period 2007-2011. These became negatively correlated after the first bank
bailoutspointing at a perception of risk transferand then exhibited a significant positive correlation, suggesting that the market was concerned about a feedback loop. Gennaioli et al. (2014) among other things find that bank holdings increase during crises,
when the expected return is high, and that much of this increase is due to the large, toobig-too-fail banks, which fits well with our banking bailout story. At the bank level, the
fact that bond holdings correlate negatively with subsequent lending during sovereign
defaults demonstrates the link between bank balance sheets and sovereign distress.
Our research leaves open a number of fascinating questions. First, we have assumed
that the bailouts take the fiscal route. As observed recently in many countries, the Central
Bank may participate in the bailout, perhaps risking inflation and devaluation. Second,
we have assumed that sovereign defaults are not strategic (the government defaults only
if it cannot repay). If defaults are strategic, domestic exposure choices by domestic banks
51

influence the incentives to default (the government is less likely to default if its debt is
held domestically), opening up the possibility of complex strategic interactions between
banks and sovereigns, and conferring a benefit (disciplining the government) upon debt
re-nationalization. Finally, further research should be devoted to the governance of the
banking union, and in particular to the interactions between prudential and fiscal integrations: should the union be committed to solidarity? Should a country bear the first losses
when one of its banks defaults? We leave these and other questions for future research.

52

References
[1] Acharya, V., Drechsler, I., and P. Schnabl (2013) A Pyrrhic Victory? Bank Bailouts
and Sovereign Credit Risk, Journal of Finance, forthcoming.
[2] Acharya, V. and S. Steffen (2013) The Greatest Carry Trade Ever? Understanding
Eurozone Bank Risks, mimeo, NYU.
[3] Acharya, V. and T. Yorulmazer (2007), Too many to fail - an analysis of timeinconsistency in bank closure policies, Journal of Financial Intermediation, 16(1): 1-31.
[4] Bianchi, J. (2013) Efficient Bailouts?, mimeo University of Wisconsin.
[5] Bocola, L. (2014) The Pass-Through of Sovereign Risk, mimeo Northwestern University and Minneapolis Fed.
[6] Broner, F., Erce, A., Martin, A. and J. Ventura (2013) Sovereign Debt Markets in
Turbulent Times: Creditor Discrimination and Crowding-out Effects, mimeo, U.
Pompeu Fabra.
[7] Broner, F. and J. Ventura (2011) Globalization and Risk Sharing, Review of Economic
Studies, 78: 49-82.
[8] Bolton, P. and O. Jeanne (2011) Sovereign Default Risk and Bank Fragility in Financially Integrated Economies, IMF Economic Review, 59: 162-194.
[9] Chari, V.V. and P. Kehoe (2013) Bailouts, Time Inconsistency, and Optimal Regulation, NBER Working Papers 19192.
[10] Chari, V.V, Dovis, A. and P. Kehoe (2014) On the Optimality of Financial Repression, mimeo Minneapolis Fed.
[11] Calvo, G. (1988) Servicing the Public Debt: The Role of Expectations, American
Economic Review, 647-661.
[12] Cooper, R. and Nikolov (2013) Government Debt and Banking Fragility: The
Spreading of Strategic Uncertainty, mimeo, Penn State University and ECB.
[13] Diamond, D., and P. Dybvig (1983) Bank Runs, Deposit Insurance and Liquidity,
Journal of Political Economy, 91: 401-19.
[14] Diamond, D., and R. Rajan (2012) Illiquid Banks, Financial Stability, and Interest
Rate Policy, Journal of Political Economy, 120(3), 552-591.
53

[15] Farhi, E. and J. Tirole (2012) Collective Moral Hazard, Maturity Mismatch, and Systemic Bailouts, American Economic Review, 102(1): 60-93.
[16] Gennaioli, N., Martin, A. and S. Rossi (2014) Sovereign Default, Domestic Banks
and Financial Institutions, Journal of Finance, 69(2): 819:866.
[28] Gennaioli, N., Martin, A. and S. Rossi (2014) Banks, Government Bonds and Defaults: What Do the Data Say? mimeo.
[18] Gennaioli, N., Shleifer, A. and R. Vishny (2012) Neglected Risk, Financial Innovation, and Financial Fragility, Journal of Financial Economics, 104: 452-468.
[19] Holmstrm, B. and J. Tirole (1997) Financial Intermediation, Loanable Funds, and
the Real Sector, Quarterly Journal of Economics, 112, 663691.
[20] Keister, T. (2014) Bailouts and Financial Fragility, Departmental Working Papers
201401, Rutgers University, Department of Economics.
[21] Lagarde, C. (2012), Opening Remarks to IMF/CFP Policy Roundtable on the Future of
Financial Regulation, International Monetary Fund, Washington DC, April 17.
[22] Mengus, E. (2013a) Honoring Sovereign Debt or Bailing Out Domestic Residents?
A Theory of Internal Cost of Default, mimeo, TSE.
[23] Mengus, E. (2013b) International Bailouts: Why Did Banks Collective Bet Lead
Europe to Rescue Greece? mimeo, TSE.
[24] Nosal, J. and G. Ordonez (2014) Uncertainty as Commitment, mimeo, Columbia
University and University of Pennsylvania.
[25] Philippon, T. and V. Skreta (2012) Optimal Interventions in Markets with Adverse
Selection, American Economic Review, 102(1), 1-28.
[26] Rancire, R., A. Tornell, and F. Westermann (2008) Systemic Crises and Growth,
The Quarterly Journal of Economics, 123(1): 359-406.
[27] Reinhart, C. and K. Rogoff (2009) This Time Is Different: Eight Centuries of Financial
Folly, Princeton University Press.
[28] Schneider, M. and A. Tornell (2004) Balance Sheet Effects, Bailout Guarantees and
Financial Crises, Review of Economic Studies, 71: 883-913.

54

[29] Stavrakeva, V. (2013) Optimal Bank Regulation and Fiscal Capacity, mimeo London Business School.
[30] Tirole, J. (2012) Overcoming Adverse Selection: How Public Intervention Can Restore Market Functioning, American Economic Review, 102(1), 29-59.
[31] Uhlig, H. (2014) Sovereign Default Risk and Banks in a Monetary Union, mimeo,
U. of Chicago.
[32] Woodford, M. (1990) Public Debt as Private Liquidity, American Economic Review,
80(2): 382-88.

Appendix: Credit Crunch Foundations of the Social Welfare Function

We sketch the foundations of the welfare function, following Homlstrm and Tirole (1997).
At date 1, the bank can make an investment in knowledge/staff so as to be able to invest
in a mass I (s) of firms at date 1. These firms enter in a relationship with the bank at date 1;
from then on, they share available resources in coalition with the banks. At date 2, firms
succeed (return r (s) per firm) or fail (return 0). Success is guaranteed if the firm managers
as well as the workers in the firm do not shirk. Otherwise success accrues with probability 0. Shirking for a firm manager brings benefit F (s), and shirking for a firm worker
W
brings benefit W (s). Therefore incentive payments 1F (s) = F (s) and W
1 (s) = (s)
per firm are required to discipline the firm manager and the workers. For simplicity, we
assume that workers and firms are cashless. Banking entrepreneurs can divert their share
of the return 1 (s) = r (s) F (s) W (s) on the project of each firm so that they cannot
borrow.
In Section 5.1, we relax the assumption that banking entrepreneurs cannot borrow.
This can be modeled as follows. Instead of assuming that banking entrepreneurs can
divert their share of the return on the project of each firm, we assume that banks need
to monitor firms. A firm succeeds if not only workers and firms do not shirk, but also
banking entrepreneurs. Shirking for a banking entrepreneur brings about a benefit B .
Banks are then able to borrow 0 (s) = r (s) ( B (s) + F (s) + W (s)) per firm that they
finance, and receive a share 1 (s) = r (s) ( F (s) + W (s)) of the return of each firm.

55